Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/architecture.png filter=lfs diff=lfs merge=lfs -text
+assets/comparison.png filter=lfs diff=lfs merge=lfs -text
+assets/motivation.png filter=lfs diff=lfs merge=lfs -text
+assets/QRCode.jpg filter=lfs diff=lfs merge=lfs -text
+assets/vis_anat.png filter=lfs diff=lfs merge=lfs -text
+assets/vis_modal.png filter=lfs diff=lfs merge=lfs -text

.github/workflows/update_space.yml

@@ -0,0 +1,28 @@
+name: Run Python script
+
+on:
+  push:
+    branches:
+      - y
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    steps:
+    - name: Checkout
+      uses: actions/checkout@v2
+
+    - name: Set up Python
+      uses: actions/setup-python@v2
+      with:
+        python-version: '3.9'
+
+    - name: Install Gradio
+      run: python -m pip install gradio
+
+    - name: Log in to Hugging Face
+      run: python -c 'import huggingface_hub; huggingface_hub.login(token="${{ secrets.hf_token }}")'
+
+    - name: Deploy to Spaces
+      run: gradio deploy

.gitignore

@@ -0,0 +1,168 @@
+data/*
+ckpt/*
+results/*
+work_dir/*
+*/__pycache__/*
+__pycache__/*
+*.pyc
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/

.gradio/certificate.pem

@@ -0,0 +1,31 @@
+-----BEGIN CERTIFICATE-----
+MIIFazCCA1OgAwIBAgIRAIIQz7DSQONZRGPgu2OCiwAwDQYJKoZIhvcNAQELBQAw
+TzELMAkGA1UEBhMCVVMxKTAnBgNVBAoTIEludGVybmV0IFNlY3VyaXR5IFJlc2Vh
+cmNoIEdyb3VwMRUwEwYDVQQDEwxJU1JHIFJvb3QgWDEwHhcNMTUwNjA0MTEwNDM4
+WhcNMzUwNjA0MTEwNDM4WjBPMQswCQYDVQQGEwJVUzEpMCcGA1UEChMgSW50ZXJu
+ZXQgU2VjdXJpdHkgUmVzZWFyY2ggR3JvdXAxFTATBgNVBAMTDElTUkcgUm9vdCBY
+MTCCAiIwDQYJKoZIhvcNAQEBBQADggIPADCCAgoCggIBAK3oJHP0FDfzm54rVygc
+h77ct984kIxuPOZXoHj3dcKi/vVqbvYATyjb3miGbESTtrFj/RQSa78f0uoxmyF+
+0TM8ukj13Xnfs7j/EvEhmkvBioZxaUpmZmyPfjxwv60pIgbz5MDmgK7iS4+3mX6U
+A5/TR5d8mUgjU+g4rk8Kb4Mu0UlXjIB0ttov0DiNewNwIRt18jA8+o+u3dpjq+sW
+T8KOEUt+zwvo/7V3LvSye0rgTBIlDHCNAymg4VMk7BPZ7hm/ELNKjD+Jo2FR3qyH
+B5T0Y3HsLuJvW5iB4YlcNHlsdu87kGJ55tukmi8mxdAQ4Q7e2RCOFvu396j3x+UC
+B5iPNgiV5+I3lg02dZ77DnKxHZu8A/lJBdiB3QW0KtZB6awBdpUKD9jf1b0SHzUv
+KBds0pjBqAlkd25HN7rOrFleaJ1/ctaJxQZBKT5ZPt0m9STJEadao0xAH0ahmbWn
+OlFuhjuefXKnEgV4We0+UXgVCwOPjdAvBbI+e0ocS3MFEvzG6uBQE3xDk3SzynTn
+jh8BCNAw1FtxNrQHusEwMFxIt4I7mKZ9YIqioymCzLq9gwQbooMDQaHWBfEbwrbw
+qHyGO0aoSCqI3Haadr8faqU9GY/rOPNk3sgrDQoo//fb4hVC1CLQJ13hef4Y53CI
+rU7m2Ys6xt0nUW7/vGT1M0NPAgMBAAGjQjBAMA4GA1UdDwEB/wQEAwIBBjAPBgNV
+HRMBAf8EBTADAQH/MB0GA1UdDgQWBBR5tFnme7bl5AFzgAiIyBpY9umbbjANBgkq
+hkiG9w0BAQsFAAOCAgEAVR9YqbyyqFDQDLHYGmkgJykIrGF1XIpu+ILlaS/V9lZL
+ubhzEFnTIZd+50xx+7LSYK05qAvqFyFWhfFQDlnrzuBZ6brJFe+GnY+EgPbk6ZGQ
+3BebYhtF8GaV0nxvwuo77x/Py9auJ/GpsMiu/X1+mvoiBOv/2X/qkSsisRcOj/KK
+NFtY2PwByVS5uCbMiogziUwthDyC3+6WVwW6LLv3xLfHTjuCvjHIInNzktHCgKQ5
+ORAzI4JMPJ+GslWYHb4phowim57iaztXOoJwTdwJx4nLCgdNbOhdjsnvzqvHu7Ur
+TkXWStAmzOVyyghqpZXjFaH3pO3JLF+l+/+sKAIuvtd7u+Nxe5AW0wdeRlN8NwdC
+jNPElpzVmbUq4JUagEiuTDkHzsxHpFKVK7q4+63SM1N95R1NbdWhscdCb+ZAJzVc
+oyi3B43njTOQ5yOf+1CceWxG1bQVs5ZufpsMljq4Ui0/1lvh+wjChP4kqKOJ2qxq
+4RgqsahDYVvTH9w7jXbyLeiNdd8XM2w9U/t7y0Ff/9yi0GE44Za4rF2LN9d11TPA
+mRGunUHBcnWEvgJBQl9nJEiU0Zsnvgc/ubhPgXRR4Xq37Z0j4r7g1SgEEzwxA57d
+emyPxgcYxn/eR44/KJ4EBs+lVDR3veyJm+kXQ99b21/+jh5Xos1AnX5iItreGCc=
+-----END CERTIFICATE-----

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

app.py

@@ -0,0 +1,399 @@
+import os
+import numpy as np
+import torch
+import SimpleITK as sitk
+import torch.nn.functional as F
+import cv2
+from PIL import Image, ImageDraw, ImageOps
+import tempfile
+import gradio as gr
+from segment_anything.build_sam3D import sam_model_registry3D
+from utils.click_method import get_next_click3D_torch_ritm, get_next_click3D_torch_2
+
+
+def build_model():
+    checkpoint_path = 'ckpt\\BoSAM.pth'
+    
+    checkpoint = torch.load(checkpoint_path, map_location='cuda', weights_only=False)
+    
+    state_dict = checkpoint['model_state_dict']
+    
+    sam_model = sam_model_registry3D['vit_b_ori'](checkpoint=None).to('cuda')
+    sam_model.load_state_dict(state_dict)
+    
+    return sam_model
+
+
+def center_crop_or_pad(image_array, target_shape=(128, 128, 128)):
+    """中心裁剪或填充图像到目标尺寸"""
+    current_shape = image_array.shape
+    
+    start = [(c - t) // 2 if c > t else 0 for c, t in zip(current_shape, target_shape)]
+    end = [s + t if c > t else c for s, t, c in zip(start, target_shape, current_shape)]
+    
+    result = np.zeros(target_shape, dtype=image_array.dtype)
+    
+    target_start = [0 if c > t else (t - c) // 2 for c, t in zip(current_shape, target_shape)]
+    target_end = [t if c > t else ts + c for ts, c, t in zip(target_start, current_shape, target_shape)]
+    
+    if all(c >= t for c, t in zip(current_shape, target_shape)):
+        cropped = image_array[
+            start[0]:start[0]+target_shape[0],
+            start[1]:start[1]+target_shape[1],
+            start[2]:start[2]+target_shape[2]
+        ]
+        return cropped
+    else:
+        source_slices = tuple(slice(0 if c <= t else s, c if c <= t else e) 
+                            for s, e, c, t in zip(start, end, current_shape, target_shape))
+        target_slices = tuple(slice(ts, te) 
+                             for ts, te in zip(target_start, target_end))
+        
+        result[target_slices] = image_array[source_slices]
+        return result
+
+
+def preprocess_image(image_path):
+    """预处理图像为128x128x128"""
+    image = sitk.ReadImage(image_path)
+    image_array = sitk.GetArrayFromImage(image)
+    
+    processed_array = center_crop_or_pad(image_array, (128, 128, 128))
+    
+    image_tensor = torch.tensor(processed_array).float().unsqueeze(0).unsqueeze(0)
+    
+    return image_tensor.to('cuda')
+
+
+def load_gt3d(image3d_path):
+    """加载并预处理GT标签为128x128x128"""
+    gt3d_path = r'examples\labels\1.3.6.1.4.1.9328.50.4.0357.nii.gz'  # 使用固定的GT
+    if not os.path.exists(gt3d_path):
+        raise FileNotFoundError(f"The file {gt3d_path} does not exist.")
+    
+    image = sitk.ReadImage(gt3d_path)
+    image_array = sitk.GetArrayFromImage(image)
+    
+    processed_array = center_crop_or_pad(image_array, (128, 128, 128))
+    
+    gt_tensor = torch.tensor(processed_array).float().unsqueeze(0).unsqueeze(0)
+    
+    return gt_tensor.to('cuda')
+
+
+def overlay_mask_on_image(image_slice, mask_slice, alpha=0.6):
+    """在图像切片上叠加掩码，增强视觉效果"""
+    # 增强对比度
+    p2, p98 = np.percentile(image_slice, (2, 98))
+    image_contrast = np.clip((image_slice - p2) / (p98 - p2), 0, 1)
+    image_contrast = (image_contrast * 255).astype(np.uint8)
+    
+    # 创建彩色图像
+    image_rgb = Image.fromarray(image_contrast).convert("RGB")
+    
+    # 应用轻微的锐化和增强
+    enhancer = ImageOps.autocontrast(image_rgb, cutoff=0.5)
+    image_rgba = enhancer.convert("RGBA")
+
+    # 创建更鲜明的掩码颜色
+    mask_image = Image.new('RGBA', image_rgba.size, (0, 0, 0, 0))
+    mask_draw = ImageDraw.Draw(mask_image)
+
+    mask = (mask_slice > 0).astype(np.uint8) * 255
+    mask_pil = Image.fromarray(mask, mode='L')
+    
+    # 使用高饱和度的蓝色
+    mask_draw.bitmap((0, 0), mask_pil, fill=(41, 128, 255, int(255 * alpha)))
+
+    # 叠加并添加轻微的发光效果
+    combined_image = Image.alpha_composite(image_rgba, mask_image)
+    
+    return combined_image
+
+
+def predict(image3D, sam_model, points=None, prev_masks=None, num_clicks=5):
+    """使用SAM模型预测掩码"""
+    sam_model.eval()
+    
+    image3D = (image3D - image3D.mean()) / image3D.std()
+    
+    gt3D = load_gt3d(None)
+    
+    if prev_masks is None:
+        prev_masks = torch.zeros_like(image3D).to('cuda')
+    
+    low_res_masks = F.interpolate(prev_masks.float(), size=(32, 32, 32))
+    
+    with torch.no_grad():
+        image_embedding = sam_model.image_encoder(image3D)
+    
+    for num_click in range(num_clicks):
+        with torch.no_grad():
+            batch_points, batch_labels = get_next_click3D_torch_2(prev_masks.to('cuda'), gt3D.to('cuda'))
+            
+            points_co = torch.cat(batch_points, dim=0).to('cuda')  
+            points_la = torch.cat(batch_labels, dim=0).to('cuda')  
+            
+            sparse_embeddings, dense_embeddings = sam_model.prompt_encoder(
+                points=[points_co, points_la],
+                boxes=None,
+                masks=low_res_masks.to('cuda'),
+            )
+            
+            low_res_masks, iou_predictions = sam_model.mask_decoder(
+                image_embeddings=image_embedding.to('cuda'),
+                image_pe=sam_model.prompt_encoder.get_dense_pe(),
+                sparse_prompt_embeddings=sparse_embeddings,
+                dense_prompt_embeddings=dense_embeddings,
+                multimask_output=False,
+            )
+            
+            prev_masks = F.interpolate(low_res_masks, size=[128, 128, 128], mode='trilinear', align_corners=False)
+            
+            medsam_seg_prob = torch.sigmoid(prev_masks)
+            medsam_seg_prob = medsam_seg_prob.cpu().numpy().squeeze()
+            medsam_seg = (medsam_seg_prob > 0.5).astype(np.uint8)
+    
+    return medsam_seg, medsam_seg_prob
+
+
+def normalize_image(image):
+    """增强图像对比度"""
+    # 使用百分位数来增强对比度
+    p2, p98 = np.percentile(image, (2, 98))
+    if p98 - p2 != 0:
+        image = np.clip((image - p2) / (p98 - p2), 0, 1)
+    else:
+        image = np.zeros_like(image)
+    image = (image * 255).astype(np.uint8)
+    return image
+
+
+def predicts(img_path, sam_model):
+    """预处理图像并预测"""
+    img = preprocess_image(img_path)
+    prediction, prediction_prob = predict(img, sam_model)
+    return prediction, prediction_prob
+
+
+def save_nifti(prediction, original_image_path):
+    """保存预测结果为NIFTI文件"""
+    original_image = sitk.ReadImage(original_image_path)
+    
+    output_image = sitk.GetImageFromArray(prediction.astype(np.uint8))
+    
+    output_image.SetDirection(original_image.GetDirection())
+    output_image.SetOrigin(original_image.GetOrigin())
+    
+    original_size = original_image.GetSize()
+    original_spacing = original_image.GetSpacing()
+    
+    new_spacing = [
+        original_spacing[0] * (original_size[0] / 128),
+        original_spacing[1] * (original_size[1] / 128),
+        original_spacing[2] * (original_size[2] / 128)
+    ]
+    output_image.SetSpacing(new_spacing)
+    
+    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".nii.gz")
+    temp_filename = temp_file.name
+    
+    sitk.WriteImage(output_image, temp_filename)
+    
+    return temp_filename
+
+
+def gr_interface(img_path, sam_model=None):
+    """增强的Gradio界面函数"""
+    if sam_model is None:
+        sam_model = build_model()
+    
+    # 显示进度信息    
+    yield None, gr.update(value="正在加载数据..."), None, None, None
+    
+    processed_img = preprocess_image(img_path)
+    
+    yield None, gr.update(value="正在分割..."), None, None, None
+    
+    prediction, prediction_prob = predicts(img_path, sam_model)
+    
+    yield None, gr.update(value="正在生成可视化..."), None, None, None
+    
+    processed_slices = []
+    combined_slices = []
+    predicted_slices = []
+    
+    nifti_file_path = save_nifti(prediction, img_path)
+    
+    # 计算中心32张切片的索引
+    start_idx = (128 - 32) // 2  # 48
+    end_idx = start_idx + 32     # 80
+    
+    for i in range(start_idx, end_idx):
+        # 处理原始图像切片
+        processed_slice = processed_img[0, 0, i].cpu().numpy()
+        processed_slices.append(normalize_image(processed_slice))
+        
+        # 处理预测掩码切片
+        mask_slice = prediction[i]
+        normalized_mask = normalize_image(mask_slice)
+        
+        # 叠加掩码到图像上 - 使用更醒目的视觉效果
+        combined_image = overlay_mask_on_image(processed_slices[-1], mask_slice)
+        combined_slices.append(combined_image)
+        
+        # 添加预测切片
+        predicted_slices.append(normalized_mask)
+    
+    yield processed_slices, gr.update(value="分割完成!"), combined_slices, predicted_slices, nifti_file_path
+
+
+# 使用示例文件路径作为常量
+DEFAULT_EXAMPLE = "examples\\1.3.6.1.4.1.9328.50.4.0327.nii.gz"
+EXAMPLES = [
+    ["examples\\1.3.6.1.4.1.9328.50.4.0327.nii.gz"],
+    ["examples\\1.3.6.1.4.1.9328.50.4.0357.nii.gz"],
+    ["examples\\1.3.6.1.4.1.9328.50.4.0477.nii.gz"],
+    ["examples\\1.3.6.1.4.1.9328.50.4.0491.nii.gz"],
+    ["examples\\1.3.6.1.4.1.9328.50.4.0708.nii.gz"],
+    ["examples\\1.3.6.1.4.1.9328.50.4.0719.nii.gz"]
+]
+
+# 自定义CSS样式以美化界面
+css = """
+body {
+    background-color: #f8fafc;
+}
+
+.container {
+    max-width: 1200px;
+    margin: 0 auto;
+}
+
+.main-title {
+    text-align: center;
+    color: #2563eb;
+    font-size: 2.5rem;
+    margin-bottom: 1rem;
+    font-weight: bold;
+    animation: fadeIn 1.5s ease-in-out;
+}
+
+.sub-title {
+    text-align: center;
+    color: #1e293b;
+    margin-bottom: 2rem;
+    animation: fadeIn 2s ease-in-out;
+}
+
+.custom-button {
+    background-color: #2563eb !important;
+    color: white !important;
+    transition: transform 0.3s, box-shadow 0.3s;
+}
+
+.custom-button:hover {
+    transform: translateY(-2px);
+    box-shadow: 0 4px 12px rgba(0, 0, 0, 0.15);
+}
+
+.gallery-item {
+    border-radius: 8px;
+    overflow: hidden;
+    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
+    transition: transform 0.3s;
+}
+
+.gallery-item:hover {
+    transform: scale(1.02);
+    box-shadow: 0 6px 12px rgba(0, 0, 0, 0.15);
+}
+
+@keyframes fadeIn {
+    from { opacity: 0; transform: translateY(20px); }
+    to { opacity: 1; transform: translateY(0); }
+}
+"""
+
+# 加载模型（全局变量）
+sam_model = build_model()
+
+# 创建更美观的Gradio界面，使用兼容的组件
+with gr.Blocks(title="3D医学影像智能分割系统", css=css) as demo:
+    gr.HTML("<h1 class='main-title'>3D医学影像智能分割系统</h1>")
+    gr.HTML("<p class='sub-title'>基于BoSAM的前沿人工智能自动分割技术，为医学影像分析提供高精度解决方案</p>")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            # 输入区域
+            gr.Markdown("### 上传/选择影像")
+            input_file = gr.File(label="上传NIfTI文件", value=DEFAULT_EXAMPLE)
+            
+            status = gr.Textbox(label="处理状态", value="准备就绪")
+            process_btn = gr.Button("开始智能分割", elem_classes=["custom-button"])
+            
+            # 示例区域
+            gr.Markdown("### 示例数据")
+            examples = gr.Examples(
+                examples=EXAMPLES,
+                inputs=[input_file]
+            )
+            
+            gr.HTML("""
+            <div style="margin-top: 2rem; padding: 1rem; background-color: rgba(16, 185, 129, 0.1); border-radius: 8px;">
+                <h3 style="color: #10b981; margin-bottom: 0.5rem;">技术亮点</h3>
+                <ul style="margin-left: 1.5rem;">
+                    <li>基于最新的Segment Anything Model (SAM) 技术</li>
+                    <li>专为3D医学影像优化的深度学习模型</li>
+                    <li>智能识别解剖结构，无需手动绘制边界</li>
+                    <li>高精度分割结果，提升诊断效率</li>
+                </ul>
+            </div>
+            """)
+
+        with gr.Column(scale=2):
+            # 输出区域
+            with gr.Row():
+                gr.Markdown("## 原始医学影像")
+                output_original = gr.Gallery(label="", show_label=False, columns=4, rows=8, height="600px", elem_classes=["gallery-item"])
+            
+            with gr.Row():
+                with gr.Column():
+                    gr.Markdown("## 分割叠加结果")
+                    output_combined = gr.Gallery(label="", show_label=False, columns=4, rows=4, height="400px", elem_classes=["gallery-item"])
+                
+                with gr.Column():
+                    gr.Markdown("## 分割掩码")
+                    output_mask = gr.Gallery(label="", show_label=False, columns=4, rows=4, height="400px", elem_classes=["gallery-item"])
+            
+            gr.Markdown("## 分割结果下载")
+            output_file = gr.File(label="下载完整3D分割结果 (NIFTI格式)")
+    
+    gr.HTML("""
+    <div style="text-align: center; margin-top: 2rem; padding: 1rem; border-top: 1px solid rgba(0, 0, 0, 0.1);">
+        <p>© 2025 3D医学影像智能分割系统 | 人工智能辅助医学影像分析平台</p>
+        <p>基于最新的BoaSAM模型，为医疗影像分析提供高精度自动分割解决方案</p>
+    </div>
+    """)
+    
+    # 处理事件
+    process_btn.click(
+        fn=gr_interface,
+        inputs=[input_file],
+        outputs=[output_original, status, output_combined, output_mask, output_file]
+    )
+    
+    examples.dataset.click(
+        fn=gr_interface,
+        inputs=[input_file],
+        outputs=[output_original, status, output_combined, output_mask, output_file]
+    )
+    
+    demo.load(
+        fn=gr_interface,
+        inputs=[input_file],
+        outputs=[output_original, status, output_combined, output_mask, output_file]
+    )
+
+if __name__ == "__main__":
+    demo.launch(debug=True, share = True)

examples/1.3.6.1.4.1.9328.50.4.0327.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17261b6a18c222ef7e7796f4bfcbef6f4c440f9135a101435e8159e0e52231a5
+size 103484205

examples/1.3.6.1.4.1.9328.50.4.0357.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:473d38a0fbec79d86882575808bdca81835898a27ce29be35e9d860c15abbaae
+size 112380358

examples/1.3.6.1.4.1.9328.50.4.0477.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bba2263ff3fa175b88033d9d4186c18802770114c9c1b01c90139daf6ef738a6
+size 26655732

examples/1.3.6.1.4.1.9328.50.4.0491.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f349f80f354c5a3a6db24d022df4ba87e431a12e080a133347f4b9278fa0d418
+size 117995983

examples/1.3.6.1.4.1.9328.50.4.0708.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f7ba2b41b1eaaff732c6a3668b1b59b77a9cdd28cf5d2156e6370fd083da26b
+size 118607138

examples/1.3.6.1.4.1.9328.50.4.0719.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b65efaa595473c9804745be94b033e06934e26b333d855c2d85d73667f6e5aa
+size 111900628

examples/labels/1.3.6.1.4.1.9328.50.4.0357.nii.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cfdee17b5ada204b3340aa4079c5883f9bb4baa723400e0e5c66945d2a5bece
+size 189731

infer.sh

@@ -0,0 +1,7 @@
+python inference.py --seed 2024\
+ -cp ./ckpt/sam_med3d_turbo.pth \
+ -tdp ./data/medical_preprocessed -nc 1 \
+ --output_dir ./results  \
+ --task_name infer_turbo  
+ #--sliding_window
+ #--save_image_and_gt

infer_sequence.py

@@ -0,0 +1,647 @@
+"""
+Run inference without label masks. Based on inference.py, and requires new click methods 
+from updated utils/click_method.py. Check the new click method details for more information.
+
+Author: Karson Chrispens
+Date: 5/15/2024
+"""
+
+import os
+import os.path as osp
+
+join = osp.join
+import argparse
+import json
+import pickle
+from collections import OrderedDict, defaultdict
+from glob import glob
+from itertools import product
+
+import numpy as np
+import SimpleITK as sitk
+import torch
+import torch.nn.functional as F
+import torchio as tio
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+
+from segment_anything import sam_model_registry
+from segment_anything.build_sam3D import sam_model_registry3D
+from segment_anything.utils.transforms3D import ResizeLongestSide3D
+from utils.click_method import (
+    get_next_click3D_torch_no_gt_naive,
+    get_next_click3D_torch_no_gt,
+)
+from utils.data_loader import Dataset_Union_ALL_Infer
+
+parser = argparse.ArgumentParser()
+parser.add_argument("-tdp", "--test_data_path", type=str, default="./data/validation")
+parser.add_argument(
+    "-cp", "--checkpoint_path", type=str, default="./ckpt/sam_med3d.pth"
+)
+parser.add_argument("--output_dir", type=str, default="./visualization")
+parser.add_argument("--task_name", type=str, default="test_amos")
+parser.add_argument("--skip_existing_pred", action="store_true", default=False)
+parser.add_argument("--save_image", action="store_true", default=True)
+parser.add_argument("--sliding_window", action="store_true", default=False)
+
+parser.add_argument("--image_size", type=int, default=256)
+parser.add_argument("--crop_size", type=int, default=128)
+parser.add_argument("--device", type=str, default="cuda")
+parser.add_argument("-mt", "--model_type", type=str, default="vit_b_ori")
+parser.add_argument("-nc", "--num_clicks", type=int, default=5)
+parser.add_argument("-pm", "--point_method", type=str, default="no_gt")
+parser.add_argument("-dt", "--data_type", type=str, default="infer")
+
+parser.add_argument("--threshold", type=int, default=0)
+parser.add_argument("--dim", type=int, default=3)
+parser.add_argument("--split_idx", type=int, default=0)
+parser.add_argument("--split_num", type=int, default=1)
+parser.add_argument("--ft2d", action="store_true", default=False)
+parser.add_argument("--seed", type=int, default=2023)
+
+args = parser.parse_args()
+
+""" parse and output_dir and task_name """
+args.output_dir = join(args.output_dir, args.task_name)
+args.pred_output_dir = join(args.output_dir, "pred")
+os.makedirs(args.output_dir, exist_ok=True)
+os.makedirs(args.pred_output_dir, exist_ok=True)
+args.save_name = join(args.output_dir, "dice.py")
+print("output_dir set to", args.output_dir)
+
+SEED = args.seed
+print("set seed as", SEED)
+torch.manual_seed(SEED)
+np.random.seed(SEED)
+
+if torch.cuda.is_available():
+    torch.cuda.init()
+
+click_methods = {
+    "no_gt": get_next_click3D_torch_no_gt,
+    "no_gt_naive": get_next_click3D_torch_no_gt_naive,
+}
+
+
+def postprocess_masks(low_res_masks, image_size, original_size):
+    ori_h, ori_w = original_size
+    masks = F.interpolate(
+        low_res_masks,
+        (image_size, image_size),
+        mode="bilinear",
+        align_corners=False,
+    )
+    if args.ft2d and ori_h < image_size and ori_w < image_size:
+        top = (image_size - ori_h) // 2
+        left = (image_size - ori_w) // 2
+        masks = masks[..., top : ori_h + top, left : ori_w + left]
+        pad = (top, left)
+    else:
+        masks = F.interpolate(
+            masks, original_size, mode="bilinear", align_corners=False
+        )
+        pad = None
+    return masks, pad
+
+
+def sam_decoder_inference(
+    target_size,
+    points_coords,
+    points_labels,
+    model,
+    image_embeddings,
+    mask_inputs=None,
+    multimask=False,
+):
+    with torch.no_grad():
+        sparse_embeddings, dense_embeddings = model.prompt_encoder(
+            points=(points_coords.to(model.device), points_labels.to(model.device)),
+            boxes=None,
+            masks=mask_inputs,
+        )
+
+        low_res_masks, iou_predictions = model.mask_decoder(
+            image_embeddings=image_embeddings,
+            image_pe=model.prompt_encoder.get_dense_pe(),
+            sparse_prompt_embeddings=sparse_embeddings,
+            dense_prompt_embeddings=dense_embeddings,
+            multimask_output=multimask,
+        )
+
+    if multimask:
+        max_values, max_indexs = torch.max(iou_predictions, dim=1)
+        max_values = max_values.unsqueeze(1)
+        iou_predictions = max_values
+        low_res = []
+        for i, idx in enumerate(max_indexs):
+            low_res.append(low_res_masks[i : i + 1, idx])
+        low_res_masks = torch.stack(low_res, 0)
+    masks = F.interpolate(
+        low_res_masks,
+        (target_size, target_size),
+        mode="bilinear",
+        align_corners=False,
+    )
+    return masks, low_res_masks, iou_predictions
+
+
+def repixel_value(arr, is_seg=False):
+    if not is_seg:
+        min_val = arr.min()
+        max_val = arr.max()
+        new_arr = (arr - min_val) / (max_val - min_val + 1e-10) * 255.0
+    return new_arr
+
+
+def random_point_sampling(mask, get_point=1):
+    if isinstance(mask, torch.Tensor):
+        mask = mask.numpy()
+    fg_coords = np.argwhere(mask == 1)[:, ::-1]
+    bg_coords = np.argwhere(mask == 0)[:, ::-1]
+
+    fg_size = len(fg_coords)
+    bg_size = len(bg_coords)
+
+    if get_point == 1:
+        if fg_size > 0:
+            index = np.random.randint(fg_size)
+            fg_coord = fg_coords[index]
+            label = 1
+        else:
+            index = np.random.randint(bg_size)
+            fg_coord = bg_coords[index]
+            label = 0
+        return torch.as_tensor([fg_coord.tolist()], dtype=torch.float), torch.as_tensor(
+            [label], dtype=torch.int
+        )
+    else:
+        num_fg = get_point // 2
+        num_bg = get_point - num_fg
+        fg_indices = np.random.choice(fg_size, size=num_fg, replace=True)
+        bg_indices = np.random.choice(bg_size, size=num_bg, replace=True)
+        fg_coords = fg_coords[fg_indices]
+        bg_coords = bg_coords[bg_indices]
+        coords = np.concatenate([fg_coords, bg_coords], axis=0)
+        labels = np.concatenate([np.ones(num_fg), np.zeros(num_bg)]).astype(int)
+        indices = np.random.permutation(get_point)
+        coords, labels = torch.as_tensor(
+            coords[indices], dtype=torch.float
+        ), torch.as_tensor(labels[indices], dtype=torch.int)
+        return coords, labels
+
+
+def finetune_model_predict2D(
+    img3D,
+    gt3D,
+    sam_model_tune,
+    target_size=256,
+    click_method="no_gt",
+    device="cuda",
+    num_clicks=1,
+    prev_masks=None,
+):
+    pred_list = []
+
+    slice_mask_list = defaultdict(list)
+
+    img3D = torch.repeat_interleave(
+        img3D, repeats=3, dim=1
+    )  # 1 channel -> 3 channel (align to RGB)
+
+    click_points = []
+    click_labels = []
+    for slice_idx in tqdm(range(img3D.size(-1)), desc="transverse slices", leave=False):
+        img2D, gt2D = repixel_value(img3D[..., slice_idx]), gt3D[..., slice_idx]
+
+        if (gt2D == 0).all():
+            empty_result = torch.zeros(list(gt3D.size()[:-1]) + [1]).to(device)
+            for iter in range(num_clicks):
+                slice_mask_list[iter].append(empty_result)
+            continue
+
+        img2D = F.interpolate(
+            img2D, (target_size, target_size), mode="bilinear", align_corners=False
+        )
+        gt2D = F.interpolate(
+            gt2D.float(), (target_size, target_size), mode="nearest"
+        ).int()
+
+        img2D, gt2D = img2D.to(device), gt2D.to(device)
+        img2D = (img2D - img2D.mean()) / img2D.std()
+
+        with torch.no_grad():
+            image_embeddings = sam_model_tune.image_encoder(img2D.float())
+
+        points_co, points_la = torch.zeros(1, 0, 2).to(device), torch.zeros(1, 0).to(
+            device
+        )
+        low_res_masks = None
+        gt_semantic_seg = gt2D[0, 0].to(device)
+        true_masks = gt_semantic_seg > 0
+        for iter in range(num_clicks):
+            if low_res_masks == None:
+                pred_masks = torch.zeros_like(true_masks).to(device)
+            else:
+                pred_masks = (prev_masks[0, 0] > 0.0).to(device)
+            fn_masks = torch.logical_and(true_masks, torch.logical_not(pred_masks))
+            fp_masks = torch.logical_and(torch.logical_not(true_masks), pred_masks)
+            mask_to_sample = torch.logical_or(fn_masks, fp_masks)
+            new_points_co, _ = random_point_sampling(mask_to_sample.cpu(), get_point=1)
+            new_points_la = (
+                torch.Tensor([1]).to(torch.int64)
+                if (true_masks[new_points_co[0, 1].int(), new_points_co[0, 0].int()])
+                else torch.Tensor([0]).to(torch.int64)
+            )
+            new_points_co, new_points_la = new_points_co[None].to(
+                device
+            ), new_points_la[None].to(device)
+            points_co = torch.cat([points_co, new_points_co], dim=1)
+            points_la = torch.cat([points_la, new_points_la], dim=1)
+            prev_masks, low_res_masks, iou_predictions = sam_decoder_inference(
+                target_size,
+                points_co,
+                points_la,
+                sam_model_tune,
+                image_embeddings,
+                mask_inputs=low_res_masks,
+                multimask=True,
+            )
+            click_points.append(new_points_co)
+            click_labels.append(new_points_la)
+
+            slice_mask, _ = postprocess_masks(
+                low_res_masks, target_size, (gt3D.size(2), gt3D.size(3))
+            )
+            slice_mask_list[iter].append(
+                slice_mask[..., None]
+            )  # append (B, C, H, W, 1)
+
+    for iter in range(num_clicks):
+        medsam_seg = torch.cat(slice_mask_list[iter], dim=-1).cpu().numpy().squeeze()
+        medsam_seg = medsam_seg > sam_model_tune.mask_threshold
+        medsam_seg = medsam_seg.astype(np.uint8)
+
+        pred_list.append(medsam_seg)
+
+    return pred_list, click_points, click_labels
+
+
+def finetune_model_predict3D(
+    img3D,
+    sam_model_tune,
+    device="cuda",
+    click_method="no_gt",
+    num_clicks=10,
+    prev_masks=None,
+):
+    img3D = norm_transform(img3D.squeeze(dim=1))  # (N, C, W, H, D)
+    img3D = img3D.unsqueeze(dim=1)
+
+    click_points = []
+    click_labels = []
+
+    pred_list = []
+
+    if prev_masks is None:
+        prev_masks = torch.zeros_like(img3D).to(device)
+    low_res_masks = F.interpolate(
+        prev_masks.float(),
+        size=(args.crop_size // 4, args.crop_size // 4, args.crop_size // 4),
+    )
+
+    with torch.no_grad():
+        image_embedding = sam_model_tune.image_encoder(
+            img3D.to(device)
+        )  # (1, 384, 16, 16, 16)
+
+    for click_idx in range(num_clicks):
+        with torch.no_grad():
+
+            batch_points, batch_labels = click_methods[click_method](
+                prev_masks.to(device), img3D.to(device), 170
+            )  # default threshold is 170, showing that here
+
+            points_co = torch.cat(batch_points, dim=0).to(device)
+            points_la = torch.cat(batch_labels, dim=0).to(device)
+
+            click_points.append(points_co)
+            click_labels.append(points_la)
+
+            points_input = points_co
+            labels_input = points_la
+
+            sparse_embeddings, dense_embeddings = sam_model_tune.prompt_encoder(
+                points=[points_input, labels_input],
+                boxes=None,
+                masks=low_res_masks.to(device),
+            )
+            low_res_masks, _ = sam_model_tune.mask_decoder(
+                image_embeddings=image_embedding.to(device),  # (B, 384, 64, 64, 64)
+                image_pe=sam_model_tune.prompt_encoder.get_dense_pe(),  # (1, 384, 64, 64, 64)
+                sparse_prompt_embeddings=sparse_embeddings,  # (B, 2, 384)
+                dense_prompt_embeddings=dense_embeddings,  # (B, 384, 64, 64, 64)
+                multimask_output=False,
+            )
+            prev_masks = F.interpolate(
+                low_res_masks,
+                size=img3D.shape[-3:],
+                mode="trilinear",
+                align_corners=False,
+            )
+
+            medsam_seg_prob = torch.sigmoid(prev_masks)  # (B, 1, 64, 64, 64)
+            # convert prob to mask
+            medsam_seg_prob = medsam_seg_prob.cpu().numpy().squeeze()
+            medsam_seg = (medsam_seg_prob > 0.5).astype(np.uint8)
+            pred_list.append(medsam_seg)
+
+    return pred_list, click_points, click_labels
+
+
+# TODO: check if this works?
+def pad_and_crop_with_sliding_window(img3D, crop_transform, offset_mode="center"):
+    subject = tio.Subject(
+        image=tio.ScalarImage(tensor=img3D.squeeze(0)),
+    )
+    padding_params, cropping_params = crop_transform.compute_crop_or_pad(subject)
+    # cropping_params: (x_start, x_max-(x_start+roi_size), y_start, ...)
+    # padding_params: (x_left_pad, x_right_pad, y_left_pad, ...)
+    if cropping_params is None:
+        cropping_params = (0, 0, 0, 0, 0, 0)
+    if padding_params is None:
+        padding_params = (0, 0, 0, 0, 0, 0)
+    roi_shape = crop_transform.target_shape
+    vol_bound = (0, img3D.shape[2], 0, img3D.shape[3], 0, img3D.shape[4])
+    center_oob_ori_roi = (
+        cropping_params[0] - padding_params[0],
+        cropping_params[0] + roi_shape[0] - padding_params[0],
+        cropping_params[2] - padding_params[2],
+        cropping_params[2] + roi_shape[1] - padding_params[2],
+        cropping_params[4] - padding_params[4],
+        cropping_params[4] + roi_shape[2] - padding_params[4],
+    )
+    window_list = []
+    offset_dict = {
+        "rounded": list(product((-32, +32, 0), repeat=3)),
+        "center": [(0, 0, 0)],
+    }
+    for offset in offset_dict[offset_mode]:
+        # get the position in original volume~(allow out-of-bound) for current offset
+        oob_ori_roi = (
+            center_oob_ori_roi[0] + offset[0],
+            center_oob_ori_roi[1] + offset[0],
+            center_oob_ori_roi[2] + offset[1],
+            center_oob_ori_roi[3] + offset[1],
+            center_oob_ori_roi[4] + offset[2],
+            center_oob_ori_roi[5] + offset[2],
+        )
+        # get corresponing padding params based on `vol_bound`
+        padding_params = [0 for i in range(6)]
+        for idx, (ori_pos, bound) in enumerate(zip(oob_ori_roi, vol_bound)):
+            pad_val = 0
+            if idx % 2 == 0 and ori_pos < bound:  # left bound
+                pad_val = bound - ori_pos
+            if idx % 2 == 1 and ori_pos > bound:
+                pad_val = ori_pos - bound
+            padding_params[idx] = pad_val
+        # get corresponding crop params after padding
+        cropping_params = (
+            oob_ori_roi[0] + padding_params[0],
+            vol_bound[1] - oob_ori_roi[1] + padding_params[1],
+            oob_ori_roi[2] + padding_params[2],
+            vol_bound[3] - oob_ori_roi[3] + padding_params[3],
+            oob_ori_roi[4] + padding_params[4],
+            vol_bound[5] - oob_ori_roi[5] + padding_params[5],
+        )
+        # pad and crop for the original subject
+        pad_and_crop = tio.Compose(
+            [
+                tio.Pad(padding_params, padding_mode=crop_transform.padding_mode),
+                tio.Crop(cropping_params),
+            ]
+        )
+        subject_roi = pad_and_crop(subject)
+        img3D_roi = subject_roi.image.data.clone().detach().unsqueeze(1)
+
+        # collect all position information, and set correct roi for sliding-windows in
+        # todo: get correct roi window of half because of the sliding
+        windows_clip = [0 for i in range(6)]
+        for i in range(3):
+            if offset[i] < 0:
+                windows_clip[2 * i] = 0
+                windows_clip[2 * i + 1] = -(roi_shape[i] + offset[i])
+            elif offset[i] > 0:
+                windows_clip[2 * i] = roi_shape[i] - offset[i]
+                windows_clip[2 * i + 1] = 0
+        pos3D_roi = dict(
+            padding_params=padding_params,
+            cropping_params=cropping_params,
+            ori_roi=(
+                cropping_params[0] + windows_clip[0],
+                cropping_params[0]
+                + roi_shape[0]
+                - padding_params[0]
+                - padding_params[1]
+                + windows_clip[1],
+                cropping_params[2] + windows_clip[2],
+                cropping_params[2]
+                + roi_shape[1]
+                - padding_params[2]
+                - padding_params[3]
+                + windows_clip[3],
+                cropping_params[4] + windows_clip[4],
+                cropping_params[4]
+                + roi_shape[2]
+                - padding_params[4]
+                - padding_params[5]
+                + windows_clip[5],
+            ),
+            pred_roi=(
+                padding_params[0] + windows_clip[0],
+                roi_shape[0] - padding_params[1] + windows_clip[1],
+                padding_params[2] + windows_clip[2],
+                roi_shape[1] - padding_params[3] + windows_clip[3],
+                padding_params[4] + windows_clip[4],
+                roi_shape[2] - padding_params[5] + windows_clip[5],
+            ),
+        )
+        pred_roi = pos3D_roi["pred_roi"]
+
+        # if((gt3D_roi[pred_roi[0]:pred_roi[1],pred_roi[2]:pred_roi[3],pred_roi[4]:pred_roi[5]]==0).all()):
+        # print("skip empty window with offset", offset)
+        #    continue
+
+        window_list.append((img3D_roi, pos3D_roi))
+    return window_list
+
+
+def save_numpy_to_nifti(in_arr: np.array, out_path, meta_info):
+    # torchio turn 1xHxWxD -> DxWxH
+    # so we need to squeeze and transpose back to HxWxD
+    ori_arr = np.transpose(in_arr.squeeze(), (2, 1, 0))
+    out = sitk.GetImageFromArray(ori_arr)
+    sitk_meta_translator = lambda x: [float(i) for i in x]
+    out.SetOrigin(sitk_meta_translator(meta_info["origin"]))
+    out.SetDirection(sitk_meta_translator(meta_info["direction"]))
+    out.SetSpacing(sitk_meta_translator(meta_info["spacing"]))
+    sitk.WriteImage(out, out_path)
+
+
+if __name__ == "__main__":
+    all_dataset_paths = glob(join(args.test_data_path, "*", "*"))
+    all_dataset_paths = list(filter(osp.isdir, all_dataset_paths))
+    print("get", len(all_dataset_paths), "datasets")
+
+    crop_transform = tio.CropOrPad(
+        target_shape=(args.crop_size, args.crop_size, args.crop_size)
+    )
+
+    infer_transform = [
+        tio.ToCanonical(),
+    ]
+
+    test_dataset = Dataset_Union_ALL_Infer(
+        paths=all_dataset_paths,
+        data_type=args.data_type,
+        transform=tio.Compose(infer_transform),
+        split_num=args.split_num,
+        split_idx=args.split_idx,
+        pcc=False,
+        get_all_meta_info=True,
+    )
+
+    test_dataloader = DataLoader(
+        dataset=test_dataset, sampler=None, batch_size=1, shuffle=True
+    )
+
+    checkpoint_path = args.checkpoint_path
+
+    device = args.device
+    print("device:", device)
+
+    if args.dim == 3:
+        sam_model_tune = sam_model_registry3D[args.model_type](checkpoint=None).to(
+            device
+        )
+        if checkpoint_path is not None:
+            model_dict = torch.load(checkpoint_path, map_location=device)
+            state_dict = model_dict["model_state_dict"]
+            sam_model_tune.load_state_dict(state_dict)
+    else:
+        raise NotImplementedError(
+            "this scipts is designed for 3D sliding-window inference, not support other dims"
+        )
+
+    sam_trans = ResizeLongestSide3D(sam_model_tune.image_encoder.img_size)
+    norm_transform = tio.ZNormalization(masking_method=lambda x: x > 0)
+
+    for batch_data in tqdm(test_dataloader):
+        image3D, meta_info = batch_data
+        img_name = meta_info["image_path"][0]
+
+        modality = osp.basename(osp.dirname(osp.dirname(osp.dirname(img_name))))
+        dataset = osp.basename(osp.dirname(osp.dirname(img_name)))
+        vis_root = osp.join(args.pred_output_dir, modality, dataset)
+        pred_path = osp.join(
+            vis_root,
+            osp.basename(img_name).replace(
+                ".nii.gz", f"_pred{args.num_clicks-1}.nii.gz"
+            ),
+        )
+
+        """ inference """
+        if args.skip_existing_pred and osp.exists(pred_path):
+            pass  # if the pred existed, skip the inference
+        else:
+            image3D_full = image3D
+            pred3D_full_dict = {
+                click_idx: torch.zeros_like(image3D_full).numpy()
+                for click_idx in range(args.num_clicks)
+            }
+            offset_mode = "center" if (not args.sliding_window) else "rounded"
+            sliding_window_list = pad_and_crop_with_sliding_window(
+                image3D_full, crop_transform, offset_mode=offset_mode
+            )
+            for image3D, pos3D in sliding_window_list:
+                seg_mask_list, points, labels = finetune_model_predict3D(
+                    image3D,
+                    sam_model_tune,
+                    device=device,
+                    click_method=args.point_method,
+                    num_clicks=args.num_clicks,
+                    prev_masks=None,
+                )
+                ori_roi, pred_roi = pos3D["ori_roi"], pos3D["pred_roi"]
+                for idx, seg_mask in enumerate(seg_mask_list):
+                    seg_mask_roi = seg_mask[
+                        ...,
+                        pred_roi[0] : pred_roi[1],
+                        pred_roi[2] : pred_roi[3],
+                        pred_roi[4] : pred_roi[5],
+                    ]
+                    pred3D_full_dict[idx][
+                        ...,
+                        ori_roi[0] : ori_roi[1],
+                        ori_roi[2] : ori_roi[3],
+                        ori_roi[4] : ori_roi[5],
+                    ] = seg_mask_roi
+
+            os.makedirs(vis_root, exist_ok=True)
+            padding_params = sliding_window_list[-1][-1]["padding_params"]
+            cropping_params = sliding_window_list[-1][-1]["cropping_params"]
+            # print(padding_params, cropping_params)
+            point_offset = np.array(
+                [
+                    cropping_params[0] - padding_params[0],
+                    cropping_params[2] - padding_params[2],
+                    cropping_params[4] - padding_params[4],
+                ]
+            )
+            points = [p.cpu().numpy() + point_offset for p in points]
+            labels = [l.cpu().numpy() for l in labels]
+            pt_info = dict(points=points, labels=labels)
+            # print("save to", osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", "_pred.nii.gz")))
+            pt_path = osp.join(
+                vis_root, osp.basename(img_name).replace(".nii.gz", "_pt.pkl")
+            )
+            pickle.dump(pt_info, open(pt_path, "wb"))
+
+            if args.save_image:
+                save_numpy_to_nifti(
+                    image3D_full,
+                    osp.join(
+                        vis_root,
+                        osp.basename(img_name).replace(".nii.gz", f"_img.nii.gz"),
+                    ),
+                    meta_info,
+                )
+            for idx, pred3D_full in pred3D_full_dict.items():
+                save_numpy_to_nifti(
+                    pred3D_full,
+                    osp.join(
+                        vis_root,
+                        osp.basename(img_name).replace(".nii.gz", f"_pred{idx}.nii.gz"),
+                    ),
+                    meta_info,
+                )
+                radius = 2
+                for pt in points[: idx + 1]:
+                    pred3D_full[
+                        ...,
+                        pt[0, 0, 0] - radius : pt[0, 0, 0] + radius,
+                        pt[0, 0, 1] - radius : pt[0, 0, 1] + radius,
+                        pt[0, 0, 2] - radius : pt[0, 0, 2] + radius,
+                    ] = 10
+                save_numpy_to_nifti(
+                    pred3D_full,
+                    osp.join(
+                        vis_root,
+                        osp.basename(img_name).replace(
+                            ".nii.gz", f"_pred{idx}_wPt.nii.gz"
+                        ),
+                    ),
+                    meta_info,
+                )
+
+    print("Done")

infer_sequence.sh

@@ -0,0 +1,5 @@
+python infer_sequence.py --seed 2023 \
+ -tdp ./data/inference -nc 1 \
+ -cp ./work_dir/fine_tune_experimental_augmented/sam_model_latest.pth \
+ --output_dir ./results/sequence  \
+ --task_name sequence

inference.py

@@ -0,0 +1,531 @@
+import os
+import os.path as osp
+join = osp.join
+import numpy as np
+from glob import glob
+import torch
+from segment_anything.build_sam3D import sam_model_registry3D
+from segment_anything.utils.transforms3D import ResizeLongestSide3D
+from segment_anything import sam_model_registry
+from tqdm import tqdm
+import argparse
+import SimpleITK as sitk
+import torch.nn.functional as F
+from torch.utils.data import DataLoader
+import SimpleITK as sitk
+import torchio as tio
+import numpy as np
+from collections import OrderedDict, defaultdict
+import json
+import pickle
+from utils.click_method import get_next_click3D_torch_ritm, get_next_click3D_torch_2
+from utils.data_loader import Dataset_Union_ALL_Val
+from itertools import product
+
+parser = argparse.ArgumentParser()
+parser.add_argument('-tdp', '--test_data_path', type=str, default='./data/validation')
+parser.add_argument('-cp', '--checkpoint_path', type=str, default='./ckpt/sam_med3d.pth')
+parser.add_argument('--output_dir', type=str, default='./visualization')
+parser.add_argument('--task_name', type=str, default='test_amos')
+parser.add_argument('--skip_existing_pred', action='store_true', default=False)
+parser.add_argument('--save_image_and_gt', action='store_true', default=False)
+parser.add_argument('--sliding_window', action='store_true', default=False)
+
+parser.add_argument('--image_size', type=int, default=256)
+parser.add_argument('--crop_size', type=int, default=128)
+parser.add_argument('--device', type=str, default='cuda')
+parser.add_argument('-mt', '--model_type', type=str, default='vit_b_ori')
+parser.add_argument('-nc', '--num_clicks', type=int, default=5)
+parser.add_argument('-pm', '--point_method', type=str, default='default')
+parser.add_argument('-dt', '--data_type', type=str, default='Ts')
+
+parser.add_argument('--threshold', type=int, default=0)
+parser.add_argument('--dim', type=int, default=3)
+parser.add_argument('--split_idx', type=int, default=0)
+parser.add_argument('--split_num', type=int, default=1)
+parser.add_argument('--ft2d', action='store_true', default=False)
+parser.add_argument('--seed', type=int, default=2023)
+
+args = parser.parse_args()
+
+''' parse and output_dir and task_name '''
+args.output_dir = join(args.output_dir, args.task_name)
+args.pred_output_dir = join(args.output_dir, "pred")
+os.makedirs(args.output_dir, exist_ok=True)
+os.makedirs(args.pred_output_dir, exist_ok=True)
+args.save_name = join(args.output_dir, "dice.py")
+print("output_dir set to", args.output_dir)
+
+SEED = args.seed
+print("set seed as", SEED)
+torch.manual_seed(SEED)
+np.random.seed(SEED)
+
+if torch.cuda.is_available():
+    torch.cuda.init()
+
+click_methods = {
+    'default': get_next_click3D_torch_ritm,
+    'ritm': get_next_click3D_torch_ritm,
+    'random': get_next_click3D_torch_2,
+}
+
+def compute_iou(pred_mask, gt_semantic_seg):
+    in_mask = np.logical_and(gt_semantic_seg, pred_mask)
+    out_mask = np.logical_or(gt_semantic_seg, pred_mask)
+    iou = np.sum(in_mask) / np.sum(out_mask)
+    return iou
+
+def compute_dice(mask_gt, mask_pred, dtype=np.uint8):
+    volume_sum = mask_gt.sum() + mask_pred.sum()
+    if volume_sum == 0:
+        return np.NaN
+    volume_intersect = (mask_gt.astype(dtype) & mask_pred.astype(dtype)).sum()
+    return 2*volume_intersect / volume_sum
+
+def postprocess_masks(low_res_masks, image_size, original_size):
+    ori_h, ori_w = original_size
+    masks = F.interpolate(
+        low_res_masks,
+        (image_size, image_size),
+        mode="bilinear",
+        align_corners=False,
+        )
+    if args.ft2d and ori_h < image_size and ori_w < image_size:
+        top = (image_size - ori_h) // 2
+        left = (image_size - ori_w) // 2
+        masks = masks[..., top : ori_h + top, left : ori_w + left]
+        pad = (top, left)
+    else:
+        masks = F.interpolate(masks, original_size, mode="bilinear", align_corners=False)
+        pad = None 
+    return masks, pad
+
+def sam_decoder_inference(target_size, points_coords, points_labels, model, image_embeddings, mask_inputs=None, multimask = False):
+    with torch.no_grad():
+        sparse_embeddings, dense_embeddings = model.prompt_encoder(
+            points=(points_coords.to(model.device), points_labels.to(model.device)),
+            boxes=None,
+            masks=mask_inputs,
+        )
+
+        low_res_masks, iou_predictions = model.mask_decoder(
+            image_embeddings = image_embeddings,
+            image_pe = model.prompt_encoder.get_dense_pe(),
+            sparse_prompt_embeddings=sparse_embeddings,
+            dense_prompt_embeddings=dense_embeddings,
+            multimask_output=multimask,
+        )
+    
+    if multimask:
+        max_values, max_indexs = torch.max(iou_predictions, dim=1)
+        max_values = max_values.unsqueeze(1)
+        iou_predictions = max_values
+        low_res = []
+        for i, idx in enumerate(max_indexs):
+            low_res.append(low_res_masks[i:i+1, idx])
+        low_res_masks = torch.stack(low_res, 0)
+    masks = F.interpolate(low_res_masks, (target_size, target_size), mode="bilinear", align_corners=False,)
+    return masks, low_res_masks, iou_predictions
+
+def repixel_value(arr, is_seg=False):
+    if not is_seg:
+        min_val = arr.min()
+        max_val = arr.max()
+        new_arr = (arr - min_val) / (max_val - min_val + 1e-10) * 255.
+    return new_arr
+
+def random_point_sampling(mask, get_point = 1):
+    if isinstance(mask, torch.Tensor):
+        mask = mask.numpy()
+    fg_coords = np.argwhere(mask == 1)[:,::-1]
+    bg_coords = np.argwhere(mask == 0)[:,::-1]
+    
+    fg_size = len(fg_coords)
+    bg_size = len(bg_coords)
+
+    if get_point == 1:
+        if fg_size > 0:
+            index = np.random.randint(fg_size)
+            fg_coord = fg_coords[index]
+            label = 1
+        else:
+            index = np.random.randint(bg_size)
+            fg_coord = bg_coords[index]
+            label = 0
+        return torch.as_tensor([fg_coord.tolist()], dtype=torch.float), torch.as_tensor([label], dtype=torch.int)
+    else:
+        num_fg = get_point // 2
+        num_bg = get_point - num_fg
+        fg_indices = np.random.choice(fg_size, size=num_fg, replace=True)
+        bg_indices = np.random.choice(bg_size, size=num_bg, replace=True)
+        fg_coords = fg_coords[fg_indices]
+        bg_coords = bg_coords[bg_indices]
+        coords = np.concatenate([fg_coords, bg_coords], axis=0)
+        labels = np.concatenate([np.ones(num_fg), np.zeros(num_bg)]).astype(int)
+        indices = np.random.permutation(get_point)
+        coords, labels = torch.as_tensor(coords[indices], dtype=torch.float), torch.as_tensor(labels[indices], dtype=torch.int)
+        return coords, labels
+
+
+def finetune_model_predict2D(img3D, gt3D, sam_model_tune, target_size=256, click_method='random', device='cuda', num_clicks=1, prev_masks=None):
+    pred_list = []
+
+    slice_mask_list = defaultdict(list)
+
+    img3D = torch.repeat_interleave(img3D, repeats=3, dim=1) # 1 channel -> 3 channel (align to RGB)
+    
+    click_points = []
+    click_labels = []
+    for slice_idx in tqdm(range(img3D.size(-1)), desc="transverse slices", leave=False):
+        img2D, gt2D = repixel_value(img3D[..., slice_idx]), gt3D[..., slice_idx]
+
+        if (gt2D==0).all():
+            empty_result = torch.zeros(list(gt3D.size()[:-1])+[1]).to(device)
+            for iter in range(num_clicks):
+                slice_mask_list[iter].append(empty_result)
+            continue
+
+        img2D = F.interpolate(img2D, (target_size, target_size), mode="bilinear", align_corners=False)
+        gt2D = F.interpolate(gt2D.float(), (target_size, target_size), mode="nearest").int()
+        
+        img2D, gt2D = img2D.to(device), gt2D.to(device)
+        img2D = (img2D - img2D.mean()) / img2D.std()
+
+        with torch.no_grad():
+            image_embeddings = sam_model_tune.image_encoder(img2D.float())
+
+        points_co, points_la = torch.zeros(1,0,2).to(device), torch.zeros(1,0).to(device)
+        low_res_masks = None
+        gt_semantic_seg = gt2D[0, 0].to(device)
+        true_masks = (gt_semantic_seg > 0)
+        for iter in range(num_clicks):
+            if(low_res_masks==None):
+                pred_masks = torch.zeros_like(true_masks).to(device)
+            else:
+                pred_masks = (prev_masks[0, 0] > 0.0).to(device) 
+            fn_masks = torch.logical_and(true_masks, torch.logical_not(pred_masks))
+            fp_masks = torch.logical_and(torch.logical_not(true_masks), pred_masks)
+            mask_to_sample = torch.logical_or(fn_masks, fp_masks)
+            new_points_co, _ = random_point_sampling(mask_to_sample.cpu(), get_point=1)
+            new_points_la = torch.Tensor([1]).to(torch.int64) if(true_masks[new_points_co[0,1].int(), new_points_co[0,0].int()]) else torch.Tensor([0]).to(torch.int64)
+            new_points_co, new_points_la = new_points_co[None].to(device), new_points_la[None].to(device)
+            points_co = torch.cat([points_co, new_points_co],dim=1)
+            points_la = torch.cat([points_la, new_points_la],dim=1)
+            prev_masks, low_res_masks, iou_predictions = sam_decoder_inference(
+                target_size, points_co, points_la, sam_model_tune, image_embeddings, 
+                mask_inputs = low_res_masks, multimask = True)
+            click_points.append(new_points_co)
+            click_labels.append(new_points_la)
+            
+            slice_mask, _ = postprocess_masks(low_res_masks, target_size, (gt3D.size(2), gt3D.size(3)))
+            slice_mask_list[iter].append(slice_mask[..., None]) # append (B, C, H, W, 1)
+        
+    for iter in range(num_clicks):
+        medsam_seg = torch.cat(slice_mask_list[iter], dim=-1).cpu().numpy().squeeze()
+        medsam_seg = medsam_seg > sam_model_tune.mask_threshold
+        medsam_seg = medsam_seg.astype(np.uint8)
+
+        pred_list.append(medsam_seg) 
+
+    return pred_list, click_points, click_labels
+
+
+def finetune_model_predict3D(img3D, gt3D, sam_model_tune, device='cuda', click_method='random', num_clicks=10, prev_masks=None):
+    img3D = norm_transform(img3D.squeeze(dim=1)) # (N, C, W, H, D)
+    img3D = img3D.unsqueeze(dim=1)
+
+    click_points = []
+    click_labels = []
+
+    pred_list = []
+
+    if prev_masks is None:
+        prev_masks = torch.zeros_like(gt3D).to(device)
+    low_res_masks = F.interpolate(prev_masks.float(), size=(args.crop_size//4,args.crop_size//4,args.crop_size//4))
+
+    with torch.no_grad():
+        image_embedding = sam_model_tune.image_encoder(img3D.to(device)) # (1, 384, 16, 16, 16)
+
+    for click_idx in range(num_clicks):
+        with torch.no_grad():
+            if(click_idx>1):
+                click_method = "random"
+            batch_points, batch_labels = click_methods[click_method](prev_masks.to(device), gt3D.to(device))
+
+            points_co = torch.cat(batch_points, dim=0).to(device)  
+            points_la = torch.cat(batch_labels, dim=0).to(device)  
+
+            click_points.append(points_co)
+            click_labels.append(points_la)
+
+            points_input = points_co
+            labels_input = points_la
+
+            sparse_embeddings, dense_embeddings = sam_model_tune.prompt_encoder(
+                points=[points_input, labels_input],
+                boxes=None,
+                masks=low_res_masks.to(device),
+            )
+            low_res_masks, _ = sam_model_tune.mask_decoder(
+                image_embeddings=image_embedding.to(device), # (B, 384, 64, 64, 64)
+                image_pe=sam_model_tune.prompt_encoder.get_dense_pe(), # (1, 384, 64, 64, 64)
+                sparse_prompt_embeddings=sparse_embeddings, # (B, 2, 384)
+                dense_prompt_embeddings=dense_embeddings, # (B, 384, 64, 64, 64)
+                multimask_output=False,
+                )
+            prev_masks = F.interpolate(low_res_masks, size=gt3D.shape[-3:], mode='trilinear', align_corners=False)
+
+            medsam_seg_prob = torch.sigmoid(prev_masks)  # (B, 1, 64, 64, 64)
+            # convert prob to mask
+            medsam_seg_prob = medsam_seg_prob.cpu().numpy().squeeze()
+            medsam_seg = (medsam_seg_prob > 0.5).astype(np.uint8)
+            pred_list.append(medsam_seg)
+
+    return pred_list, click_points, click_labels
+
+
+def pad_and_crop_with_sliding_window(img3D, gt3D, crop_transform, offset_mode="center"):
+    subject = tio.Subject(
+        image = tio.ScalarImage(tensor=img3D.squeeze(0)),
+        label = tio.LabelMap(tensor=gt3D.squeeze(0)),
+    )
+    padding_params, cropping_params = crop_transform.compute_crop_or_pad(subject)
+    # cropping_params: (x_start, x_max-(x_start+roi_size), y_start, ...)
+    # padding_params: (x_left_pad, x_right_pad, y_left_pad, ...)
+    if(cropping_params is None): cropping_params = (0,0,0,0,0,0)
+    if(padding_params is None): padding_params = (0,0,0,0,0,0)
+    roi_shape = crop_transform.target_shape
+    vol_bound = (0, img3D.shape[2], 0, img3D.shape[3], 0, img3D.shape[4])
+    center_oob_ori_roi=(
+        cropping_params[0]-padding_params[0], cropping_params[0]+roi_shape[0]-padding_params[0],
+        cropping_params[2]-padding_params[2], cropping_params[2]+roi_shape[1]-padding_params[2],
+        cropping_params[4]-padding_params[4], cropping_params[4]+roi_shape[2]-padding_params[4],
+    )
+    window_list = []
+    offset_dict = {
+        "rounded": list(product((-32,+32,0), repeat=3)),
+        "center": [(0,0,0)],
+    }
+    for offset in offset_dict[offset_mode]:
+        # get the position in original volume~(allow out-of-bound) for current offset
+        oob_ori_roi = (
+            center_oob_ori_roi[0]+offset[0], center_oob_ori_roi[1]+offset[0],
+            center_oob_ori_roi[2]+offset[1], center_oob_ori_roi[3]+offset[1],
+            center_oob_ori_roi[4]+offset[2], center_oob_ori_roi[5]+offset[2],
+        )
+        # get corresponing padding params based on `vol_bound`
+        padding_params = [0 for i in range(6)]
+        for idx, (ori_pos, bound) in enumerate(zip(oob_ori_roi, vol_bound)):
+            pad_val = 0
+            if(idx%2==0 and ori_pos<bound): # left bound
+                pad_val = bound-ori_pos
+            if(idx%2==1 and ori_pos>bound):
+                pad_val = ori_pos-bound
+            padding_params[idx] = pad_val
+        # get corresponding crop params after padding
+        cropping_params = (
+            oob_ori_roi[0]+padding_params[0], vol_bound[1]-oob_ori_roi[1]+padding_params[1],
+            oob_ori_roi[2]+padding_params[2], vol_bound[3]-oob_ori_roi[3]+padding_params[3],
+            oob_ori_roi[4]+padding_params[4], vol_bound[5]-oob_ori_roi[5]+padding_params[5],
+        )
+        # pad and crop for the original subject
+        pad_and_crop = tio.Compose([
+            tio.Pad(padding_params, padding_mode=crop_transform.padding_mode),
+            tio.Crop(cropping_params),
+        ])
+        subject_roi = pad_and_crop(subject)  
+        img3D_roi, gt3D_roi = subject_roi.image.data.clone().detach().unsqueeze(1), subject_roi.label.data.clone().detach().unsqueeze(1)
+
+        # collect all position information, and set correct roi for sliding-windows in 
+        # todo: get correct roi window of half because of the sliding 
+        windows_clip = [0 for i in range(6)]
+        for i in range(3):
+            if(offset[i]<0):
+                windows_clip[2*i] = 0
+                windows_clip[2*i+1] = -(roi_shape[i]+offset[i])
+            elif(offset[i]>0):
+                windows_clip[2*i] = roi_shape[i]-offset[i]
+                windows_clip[2*i+1] = 0
+        pos3D_roi = dict(
+            padding_params=padding_params, cropping_params=cropping_params, 
+            ori_roi=(
+                cropping_params[0]+windows_clip[0], cropping_params[0]+roi_shape[0]-padding_params[0]-padding_params[1]+windows_clip[1],
+                cropping_params[2]+windows_clip[2], cropping_params[2]+roi_shape[1]-padding_params[2]-padding_params[3]+windows_clip[3],
+                cropping_params[4]+windows_clip[4], cropping_params[4]+roi_shape[2]-padding_params[4]-padding_params[5]+windows_clip[5],
+            ),
+            pred_roi=(
+                padding_params[0]+windows_clip[0], roi_shape[0]-padding_params[1]+windows_clip[1],
+                padding_params[2]+windows_clip[2], roi_shape[1]-padding_params[3]+windows_clip[3],
+                padding_params[4]+windows_clip[4], roi_shape[2]-padding_params[5]+windows_clip[5],
+            ))
+        pred_roi = pos3D_roi["pred_roi"]
+        
+        #if((gt3D_roi[pred_roi[0]:pred_roi[1],pred_roi[2]:pred_roi[3],pred_roi[4]:pred_roi[5]]==0).all()):
+            #print("skip empty window with offset", offset)
+        #    continue
+
+        window_list.append((img3D_roi, gt3D_roi, pos3D_roi))
+    return window_list
+
+def save_numpy_to_nifti(in_arr: np.array, out_path, meta_info):
+    # torchio turn 1xHxWxD -> DxWxH
+    # so we need to squeeze and transpose back to HxWxD
+    ori_arr = np.transpose(in_arr.squeeze(), (2, 1, 0))
+    out = sitk.GetImageFromArray(ori_arr)
+    sitk_meta_translator = lambda x: [float(i) for i in x]
+    out.SetOrigin(sitk_meta_translator(meta_info["origin"]))
+    out.SetDirection(sitk_meta_translator(meta_info["direction"]))
+    out.SetSpacing(sitk_meta_translator(meta_info["spacing"]))
+    sitk.WriteImage(out, out_path)
+
+
+if __name__ == "__main__":    
+    all_dataset_paths = glob(join(args.test_data_path, "*", "*"))
+    all_dataset_paths = list(filter(osp.isdir, all_dataset_paths))
+    print("get", len(all_dataset_paths), "datasets")
+
+    crop_transform = tio.CropOrPad(
+        mask_name='label', 
+        target_shape=(args.crop_size, args.crop_size, args.crop_size))
+    
+    infer_transform = [
+        tio.ToCanonical(),
+    ]
+
+    test_dataset = Dataset_Union_ALL_Val(
+        paths=all_dataset_paths, 
+        mode="Val", 
+        data_type=args.data_type, 
+        transform=tio.Compose(infer_transform),
+        threshold=0,
+        split_num=args.split_num,
+        split_idx=args.split_idx,
+        pcc=False,
+        get_all_meta_info=True,
+    )
+
+    test_dataloader = DataLoader(
+        dataset=test_dataset,
+        sampler=None,
+        batch_size=1, 
+        shuffle=True
+    )
+
+    checkpoint_path = args.checkpoint_path
+
+    device = args.device
+    print("device:", device)
+
+    if(args.dim==3):
+        sam_model_tune = sam_model_registry3D[args.model_type](checkpoint=None).to(device)
+        if checkpoint_path is not None:
+            model_dict = torch.load(checkpoint_path, map_location=device)
+            state_dict = model_dict['model_state_dict']
+            sam_model_tune.load_state_dict(state_dict)
+    else:
+        raise NotImplementedError("this scipts is designed for 3D sliding-window inference, not support other dims")
+
+    sam_trans = ResizeLongestSide3D(sam_model_tune.image_encoder.img_size)
+    norm_transform = tio.ZNormalization(masking_method=lambda x: x > 0)
+
+    all_iou_list = []
+    all_dice_list = []  
+
+    out_dice = dict()
+    out_dice_all = OrderedDict()
+
+    for batch_data in tqdm(test_dataloader):
+        image3D, gt3D, meta_info = batch_data
+        img_name = meta_info["image_path"][0]
+
+        modality = osp.basename(osp.dirname(osp.dirname(osp.dirname(img_name))))
+        dataset = osp.basename(osp.dirname(osp.dirname(img_name)))
+        vis_root = osp.join(args.pred_output_dir, modality, dataset)
+        pred_path = osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", f"_pred{args.num_clicks-1}.nii.gz"))
+        
+        ''' inference '''
+        iou_list, dice_list = [], []
+        if(args.skip_existing_pred and osp.exists(pred_path)):
+            pass # if the pred existed, skip the inference
+        else: 
+            image3D_full, gt3D_full = image3D, gt3D
+            pred3D_full_dict  = {click_idx:torch.zeros_like(gt3D_full).numpy() for click_idx in range(args.num_clicks)}
+            offset_mode = "center" if(not args.sliding_window) else "rounded"
+            sliding_window_list = pad_and_crop_with_sliding_window(image3D_full, gt3D_full, crop_transform, offset_mode=offset_mode)
+            for (image3D, gt3D, pos3D) in sliding_window_list:
+                seg_mask_list, points, labels = finetune_model_predict3D(
+                    image3D, gt3D, sam_model_tune, device=device, 
+                    click_method=args.point_method, num_clicks=args.num_clicks, 
+                    prev_masks=None)
+                ori_roi, pred_roi = pos3D["ori_roi"], pos3D["pred_roi"]
+                for idx, seg_mask in enumerate(seg_mask_list):
+                    seg_mask_roi = seg_mask[..., pred_roi[0]:pred_roi[1], pred_roi[2]:pred_roi[3], pred_roi[4]:pred_roi[5]]
+                    pred3D_full_dict[idx][..., ori_roi[0]:ori_roi[1], ori_roi[2]:ori_roi[3], ori_roi[4]:ori_roi[5]] = seg_mask_roi
+
+            os.makedirs(vis_root, exist_ok=True)
+            padding_params = sliding_window_list[-1][-1]["padding_params"]
+            cropping_params = sliding_window_list[-1][-1]["cropping_params"]
+            # print(padding_params, cropping_params)
+            point_offset = np.array([cropping_params[0]-padding_params[0], cropping_params[2]-padding_params[2], cropping_params[4]-padding_params[4]])
+            points = [p.cpu().numpy()+point_offset for p in points]
+            labels = [l.cpu().numpy() for l in labels]
+            pt_info = dict(points=points, labels=labels)
+            # print("save to", osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", "_pred.nii.gz")))
+            pt_path=osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", "_pt.pkl"))
+            pickle.dump(pt_info, open(pt_path, "wb"))              
+
+            if(args.save_image_and_gt):
+                save_numpy_to_nifti(image3D_full, osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", f"_img.nii.gz")), meta_info)
+                save_numpy_to_nifti(gt3D_full, osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", f"_gt.nii.gz")), meta_info)
+            for idx, pred3D_full in pred3D_full_dict.items():
+                save_numpy_to_nifti(pred3D_full, osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", f"_pred{idx}.nii.gz")), meta_info)
+                radius = 2
+                for pt in points[:idx+1]:
+                    pred3D_full[..., pt[0,0,0]-radius:pt[0,0,0]+radius, pt[0,0,1]-radius:pt[0,0,1]+radius, pt[0,0,2]-radius:pt[0,0,2]+radius] = 10
+                save_numpy_to_nifti(pred3D_full, osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", f"_pred{idx}_wPt.nii.gz")), meta_info)
+                
+        ''' metric computation '''
+        for click_idx in range(args.num_clicks):
+            reorient_tensor = lambda in_arr : np.transpose(in_arr.squeeze().detach().cpu().numpy(), (2, 1, 0))
+            curr_pred_path = osp.join(vis_root, osp.basename(img_name).replace(".nii.gz", f"_pred{click_idx}.nii.gz"))
+            medsam_seg = sitk.GetArrayFromImage(sitk.ReadImage(curr_pred_path))
+            iou_list.append(round(compute_iou(medsam_seg, reorient_tensor(gt3D_full)), 4))
+            dice_list.append(round(compute_dice(reorient_tensor(gt3D_full), medsam_seg), 4))
+
+        per_iou = max(iou_list)
+        all_iou_list.append(per_iou)
+        all_dice_list.append(max(dice_list))
+        print(dice_list)
+        out_dice[img_name] = max(dice_list)
+        cur_dice_dict = OrderedDict()
+        for i, dice in enumerate(dice_list):
+            cur_dice_dict[f'{i}'] = dice
+        out_dice_all[img_name] = cur_dice_dict
+
+    print('Mean IoU : ', sum(all_iou_list)/len(all_iou_list))
+    print('Mean Dice: ', sum(all_dice_list)/len(all_dice_list))
+
+    final_dice_dict = OrderedDict()
+    for k, v in out_dice_all.items():
+        organ = k.split('/')[-4]
+        final_dice_dict[organ] = OrderedDict()
+    for k, v in out_dice_all.items():
+        organ = k.split('/')[-4]
+        final_dice_dict[organ][k] = v
+
+    if(args.split_num>1):
+        args.save_name = args.save_name.replace('.py', f'_s{args.split_num}i{args.split_idx}.py')
+
+    print("Save to", args.save_name)
+    with open(args.save_name, 'w') as f:
+        f.writelines(f'# mean dice: \t{np.mean(all_dice_list)}\n')
+        f.writelines('dice_Ts = {')
+        for k, v in out_dice.items():
+            f.writelines(f'\'{str(k[0])}\': {v},\n')
+        f.writelines('}')
+
+    with open(args.save_name.replace('.py', '.json'), 'w') as f:
+        json.dump(final_dice_dict, f, indent=4)
+
+    print("Done")

medim_infer.py

@@ -0,0 +1,294 @@
+# -*- encoding: utf-8 -*-
+'''
+@File    :   infer_with_medim.py
+@Time    :   2024/09/08 11:31:02
+@Author  :   Haoyu Wang 
+@Contact :   small_dark@sina.com
+@Brief   :   Example code for inference with MedIM
+'''
+
+import medim
+import torch
+import numpy as np
+import torch.nn.functional as F
+import torchio as tio
+import os.path as osp
+import os
+from torchio.data.io import sitk_to_nib
+import SimpleITK as sitk
+
+
+def random_sample_next_click(prev_mask, gt_mask):
+    """
+    Randomly sample one click from ground-truth mask and previous seg mask
+
+    Arguements:
+        prev_mask: (torch.Tensor) [H,W,D] previous mask that SAM-Med3D predict
+        gt_mask: (torch.Tensor) [H,W,D] ground-truth mask for this image
+    """
+    prev_mask = prev_mask > 0
+    true_masks = gt_mask > 0
+
+    if (not true_masks.any()):
+        raise ValueError("Cannot find true value in the ground-truth!")
+
+    fn_masks = torch.logical_and(true_masks, torch.logical_not(prev_mask))
+    fp_masks = torch.logical_and(torch.logical_not(true_masks), prev_mask)
+
+    to_point_mask = torch.logical_or(fn_masks, fp_masks)
+
+    all_points = torch.argwhere(to_point_mask)
+    point = all_points[np.random.randint(len(all_points))]
+
+    if fn_masks[point[0], point[1], point[2]]:
+        is_positive = True
+    else:
+        is_positive = False
+
+    sampled_point = point.clone().detach().reshape(1, 1, 3)
+    sampled_label = torch.tensor([
+        int(is_positive),
+    ]).reshape(1, 1)
+
+    return sampled_point, sampled_label
+
+
+def sam_model_infer(model,
+                    roi_image,
+                    prompt_generator=random_sample_next_click,
+                    roi_gt=None,
+                    prev_low_res_mask=None):
+    '''
+    Inference for SAM-Med3D, inputs prompt points with its labels (positive/negative for each points)
+
+    # roi_image: (torch.Tensor) cropped image, shape [1,1,128,128,128]
+    # prompt_points_and_labels: (Tuple(torch.Tensor, torch.Tensor))
+    '''
+
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    print("using device", device)
+    model = model.to(device)
+
+    with torch.no_grad():
+        input_tensor = roi_image.to(device)
+        image_embeddings = model.image_encoder(input_tensor)
+
+        points_coords, points_labels = torch.zeros(1, 0,
+                                                   3).to(device), torch.zeros(
+                                                       1, 0).to(device)
+        new_points_co, new_points_la = torch.Tensor(
+            [[[64, 64, 64]]]).to(device), torch.Tensor([[1]]).to(torch.int64)
+        if (roi_gt is not None):
+            prev_low_res_mask = prev_low_res_mask if (
+                prev_low_res_mask is not None) else torch.zeros(
+                    1, 1, roi_image.shape[2] // 4, roi_image.shape[3] //
+                    4, roi_image.shape[4] // 4)
+            new_points_co, new_points_la = prompt_generator(
+                torch.zeros_like(roi_image)[0, 0], roi_gt[0, 0])
+            new_points_co, new_points_la = new_points_co.to(
+                device), new_points_la.to(device)
+        points_coords = torch.cat([points_coords, new_points_co], dim=1)
+        points_labels = torch.cat([points_labels, new_points_la], dim=1)
+
+        sparse_embeddings, dense_embeddings = model.prompt_encoder(
+            points=[points_coords, points_labels],
+            boxes=None,  # we currently not support bbox prompt
+            masks=prev_low_res_mask.to(device),
+            # masks=None,
+        )
+
+        low_res_masks, _ = model.mask_decoder(
+            image_embeddings=image_embeddings,  # (1, 384, 8, 8, 8)
+            image_pe=model.prompt_encoder.get_dense_pe(),  # (1, 384, 8, 8, 8)
+            sparse_prompt_embeddings=sparse_embeddings,  # (1, 2, 384)
+            dense_prompt_embeddings=dense_embeddings,  # (1, 384, 8, 8, 8)
+        )
+
+        prev_mask = F.interpolate(low_res_masks,
+                                  size=roi_image.shape[-3:],
+                                  mode='trilinear',
+                                  align_corners=False)
+
+    # convert prob to mask
+    medsam_seg_prob = torch.sigmoid(prev_mask)  # (1, 1, 64, 64, 64)
+    medsam_seg_prob = medsam_seg_prob.cpu().numpy().squeeze()
+    medsam_seg_mask = (medsam_seg_prob > 0.5).astype(np.uint8)
+
+    return medsam_seg_mask
+
+
+def resample_nii(input_path: str,
+                 output_path: str,
+                 target_spacing: tuple = (1.5, 1.5, 1.5),
+                 n=None,
+                 reference_image=None,
+                 mode="linear"):
+    """
+    Resample a nii.gz file to a specified spacing using torchio.
+
+    Parameters:
+    - input_path: Path to the input .nii.gz file.
+    - output_path: Path to save the resampled .nii.gz file.
+    - target_spacing: Desired spacing for resampling. Default is (1.5, 1.5, 1.5).
+    """
+    # Load the nii.gz file using torchio
+    subject = tio.Subject(img=tio.ScalarImage(input_path))
+    resampler = tio.Resample(target=target_spacing, image_interpolation=mode)
+    resampled_subject = resampler(subject)
+
+    if (n != None):
+        image = resampled_subject.img
+        tensor_data = image.data
+        if (isinstance(n, int)):
+            n = [n]
+        for ni in n:
+            tensor_data[tensor_data == ni] = -1
+        tensor_data[tensor_data != -1] = 0
+        tensor_data[tensor_data != 0] = 1
+        save_image = tio.ScalarImage(tensor=tensor_data, affine=image.affine)
+        reference_size = reference_image.shape[
+            1:]  # omitting the channel dimension
+        cropper_or_padder = tio.CropOrPad(reference_size)
+        save_image = cropper_or_padder(save_image)
+    else:
+        save_image = resampled_subject.img
+
+    save_image.save(output_path)
+
+
+def read_data_from_nii(img_path, gt_path):
+    sitk_image = sitk.ReadImage(img_path)
+    sitk_label = sitk.ReadImage(gt_path)
+
+    if sitk_image.GetOrigin() != sitk_label.GetOrigin():
+        sitk_image.SetOrigin(sitk_label.GetOrigin())
+    if sitk_image.GetDirection() != sitk_label.GetDirection():
+        sitk_image.SetDirection(sitk_label.GetDirection())
+
+    sitk_image_arr, _ = sitk_to_nib(sitk_image)
+    sitk_label_arr, _ = sitk_to_nib(sitk_label)
+
+    subject = tio.Subject(
+        image=tio.ScalarImage(tensor=sitk_image_arr),
+        label=tio.LabelMap(tensor=sitk_label_arr),
+    )
+    crop_transform = tio.CropOrPad(mask_name='label',
+                                   target_shape=(128, 128, 128))
+    padding_params, cropping_params = crop_transform.compute_crop_or_pad(
+        subject)
+    if (cropping_params is None): cropping_params = (0, 0, 0, 0, 0, 0)
+    if (padding_params is None): padding_params = (0, 0, 0, 0, 0, 0)
+
+    infer_transform = tio.Compose([
+        crop_transform,
+        tio.ZNormalization(masking_method=lambda x: x > 0),
+    ])
+    subject_roi = infer_transform(subject)
+
+    img3D_roi, gt3D_roi = subject_roi.image.data.clone().detach().unsqueeze(
+        1), subject_roi.label.data.clone().detach().unsqueeze(1)
+    ori_roi_offset = (
+        cropping_params[0],
+        cropping_params[0] + 128 - padding_params[0] - padding_params[1],
+        cropping_params[2],
+        cropping_params[2] + 128 - padding_params[2] - padding_params[3],
+        cropping_params[4],
+        cropping_params[4] + 128 - padding_params[4] - padding_params[5],
+    )
+
+    meta_info = {
+        "image_path": img_path,
+        "image_shape": sitk_image_arr.shape[1:],
+        "origin": sitk_label.GetOrigin(),
+        "direction": sitk_label.GetDirection(),
+        "spacing": sitk_label.GetSpacing(),
+        "padding_params": padding_params,
+        "cropping_params": cropping_params,
+        "ori_roi": ori_roi_offset,
+    }
+    return (
+        img3D_roi,
+        gt3D_roi,
+        meta_info,
+    )
+
+
+def save_numpy_to_nifti(in_arr: np.array, out_path, meta_info):
+    # torchio turn 1xHxWxD -> DxWxH
+    # so we need to squeeze and transpose back to HxWxD
+    ori_arr = np.transpose(in_arr.squeeze(), (2, 1, 0))
+    out = sitk.GetImageFromArray(ori_arr)
+    sitk_meta_translator = lambda x: [float(i) for i in x]
+    out.SetOrigin(sitk_meta_translator(meta_info["origin"]))
+    out.SetDirection(sitk_meta_translator(meta_info["direction"]))
+    out.SetSpacing(sitk_meta_translator(meta_info["spacing"]))
+    sitk.WriteImage(out, out_path)
+
+
+def data_preprocess(img_path, gt_path, category_index):
+    target_img_path = osp.join(
+        osp.dirname(img_path),
+        osp.basename(img_path).replace(".nii.gz", "_resampled.nii.gz"))
+    target_gt_path = osp.join(
+        osp.dirname(gt_path),
+        osp.basename(gt_path).replace(".nii.gz", "_resampled.nii.gz"))
+    resample_nii(img_path, target_img_path)
+    resample_nii(gt_path,
+                 target_gt_path,
+                 n=category_index,
+                 reference_image=tio.ScalarImage(target_img_path),
+                 mode="nearest")
+    roi_image, roi_label, meta_info = read_data_from_nii(
+        target_img_path, target_gt_path)
+    return roi_image, roi_label, meta_info
+
+
+def data_postprocess(roi_pred, meta_info, output_path, ori_img_path):
+    os.makedirs(osp.dirname(output_path), exist_ok=True)
+    pred3D_full = np.zeros(meta_info["image_shape"])
+    padding_params = meta_info["padding_params"]
+    unpadded_pred = roi_pred[padding_params[0] : 128-padding_params[1],
+                             padding_params[2] : 128-padding_params[3],
+                             padding_params[4] : 128-padding_params[5]]
+    ori_roi = meta_info["ori_roi"]
+    pred3D_full[ori_roi[0]:ori_roi[1], ori_roi[2]:ori_roi[3],
+                ori_roi[4]:ori_roi[5]] = unpadded_pred
+
+    sitk_image = sitk.ReadImage(ori_img_path)
+    ori_meta_info = {
+        "image_path": ori_img_path,
+        "image_shape": sitk_image.GetSize(),
+        "origin": sitk_image.GetOrigin(),
+        "direction": sitk_image.GetDirection(),
+        "spacing": sitk_image.GetSpacing(),
+    }
+    pred3D_full_ori = F.interpolate(
+        torch.Tensor(pred3D_full)[None][None],
+        size=ori_meta_info["image_shape"],
+        mode='nearest').cpu().numpy().squeeze()
+    save_numpy_to_nifti(pred3D_full_ori, output_path, meta_info)
+
+
+if __name__ == "__main__":
+    ''' 1. read and pre-process your input data '''
+    img_path = "./test_data/kidney_right/AMOS/imagesVal/amos_0013.nii.gz"
+    gt_path =  "./test_data/kidney_right/AMOS/labelsVal/amos_0013.nii.gz"
+    category_index = 3  # the index of your target category in the gt annotation
+    output_dir = "./test_data/kidney_right/AMOS/pred/"
+    roi_image, roi_label, meta_info = data_preprocess(img_path, gt_path, category_index=category_index)
+    
+    ''' 2. prepare the pre-trained model with local path or huggingface url '''
+    ckpt_path = "https://huggingface.co/blueyo0/SAM-Med3D/blob/main/sam_med3d_turbo.pth"
+    # or you can use the local path like: ckpt_path = "./ckpt/sam_med3d_turbo.pth"
+    model = medim.create_model("SAM-Med3D",
+                               pretrained=True,
+                               checkpoint_path=ckpt_path)
+    
+    ''' 3. infer with the pre-trained SAM-Med3D model '''
+    roi_pred = sam_model_infer(model, roi_image, roi_gt=roi_label)
+
+    ''' 4. post-process and save the result '''
+    output_path = osp.join(output_dir, osp.basename(img_path).replace(".nii.gz", "_pred.nii.gz"))
+    data_postprocess(roi_pred, meta_info, output_path, img_path)
+
+    print("result saved to", output_path)

readme.md

@@ -0,0 +1,286 @@
+---
+title: BoSAM
+app_file: app.py
+sdk: gradio
+sdk_version: 5.25.2
+---
+# SAM-Med3D \[[Paper](https://arxiv.org/abs/2310.15161)] \[[Suppl](https://github.com/uni-medical/SAM-Med3D/blob/main/paper/SAM_Med3D_ECCV_Supplementary.pdf)\] \[[Data](https://drive.google.com/file/d/1F7lRWM5mdEKSRQtvJ8ExEyNrWIEkXc-G/view?usp=drive_link)\]
+[![x](https://img.shields.io/badge/cs.CV-2310.15161-b31b1b?logo=arxiv&logoColor=red)](https://arxiv.org/abs/2310.15161)
+[![x](https://img.shields.io/badge/WeChat-Group-green?logo=wechat)](https://github.com/uni-medical/SAM-Med3D/tree/main?tab=readme-ov-file#-discussion-group)
+
+
+The official repo of "SAM-Med3D: Towards General-purpose Segmentation Models for Volumetric Medical Images".
+
+<div align="center">
+  <img src="assets/motivation.png">
+</div>
+
+## 🔥🌻📰 News 📰🌻🔥
+- **[Challenge]** SAM-Med3D is invited as a baseline of [CVPR-MedSegFMCompetition](https://www.codabench.org/competitions/5263/) and the tutorial is [here](https://github.com/uni-medical/SAM-Med3D/tree/CVPR25_3DFM). We kindly invite you to join the challenge and build better foundation models for 3D medical image segmentation!
+- **[Examples]** SAM-Med3D is now supported in [MedIM](https://github.com/uni-medical/MedIM), you can easily get our model with one-line Python code. Our new example is in [`medim_infer.py`](https://github.com/uni-medical/SAM-Med3D/blob/main/medim_infer.py).
+- **[Data]** We have now released all labels of our training dataset SA-Med3D-140K. Due to the large volume of image data (over 1T), we are currently seeking an appropriate release method. For now, you can directly contact small_dark@sina.com to obtain it. Download Link: [Baidu Netdisk](https://pan.baidu.com/s/12Nxwd10uVZs57O8WP8Y-Hg?pwd=cv6t) and [Google Drive](https://drive.google.com/file/d/1F7lRWM5mdEKSRQtvJ8ExEyNrWIEkXc-G/view?usp=drive_link).
+- **[Paper]** SAM-Med3D is accepted as [ECCV BIC 2024 Oral](https://www.bioimagecomputing.com/program/selected-contributions/)
+- **[Model]** A newer version of finetuned SAM-Med3D named `SAM-Med3D-turbo` is released now. We fine-tuned it on 44 datasets ([list](https://github.com/uni-medical/SAM-Med3D/issues/2#issuecomment-1849002225)) to improve the performance. Hope this update can help you 🙂.
+- **[Repos]** If you are interested in computer vision, 
+we recommend checking out [OpenGVLab](https://github.com/OpenGVLab) for more exciting projects like [SAM-Med2D](https://github.com/OpenGVLab/SAM-Med2D/tree/main)!
+
+## 🌟 Highlights
+- 📚 Curated the most extensive volumetric medical dataset to date for training, boasting 143K 3D masks and 245 categories.
+- 🚤 Achieved efficient promptable segmentation, requiring 10 to 100 times fewer prompt points for satisfactory 3D outcomes.
+- 🏆 Conducted a thorough assessment of SAM-Med3D across 16 frequently used volumetric medical image segmentation datasets.
+
+## 🔗 Checkpoint
+
+ - **SAM-Med3D-turbo**: [Hugging Face](https://huggingface.co/blueyo0/SAM-Med3D/blob/main/sam_med3d_turbo.pth) | [Google Drive](https://drive.google.com/file/d/1MuqYRQKIZb4YPtEraK8zTKKpp-dUQIR9/view?usp=sharing) | [Baidu NetDisk (pwd:l6ol)](https://pan.baidu.com/s/1OEVtiDc6osG0l9HkQN4hEg?pwd=l6ol)
+
+<details>
+  <summary>More</summary>
+
+- **SAM-Med3D-base**:  
+  [Google Drive](https://drive.google.com/file/d/1PFeUjlFMAppllS9x1kAWyCYUJM9re2Ub/view?usp=drive_link) | [Baidu NetDisk (pwd: r5o3)](https://pan.baidu.com/s/18uhMXy_XO0yy3ODj66N8GQ?pwd=r5o3)  
+- **SAM-Med3D-organ**:  
+  [Google Drive](https://drive.google.com/file/d/1kKpjIwCsUWQI-mYZ2Lww9WZXuJxc3FvU/view?usp=sharing) | [Baidu NetDisk (pwd: 5t7v)](https://pan.baidu.com/s/1Dermdr-ZN8NMWELejF1p1w?pwd=5t7v)  
+- **SAM-Med3D-brain**:  
+  [Google Drive](https://drive.google.com/file/d/1otbhZs9uugSWkAbcQLLSmPB8jo5rzFL2/view?usp=sharing) | [Baidu NetDisk (pwd: yp42)](https://pan.baidu.com/s/1S2-buTga9D4Nbrt6fevo8Q?pwd=yp42)  
+
+Other checkpoints are available via their official links:  
+- [SAM](https://drive.google.com/file/d/1_U26MIJhWnWVwmI5JkGg2cd2J6MvkqU-/view?usp=drive_link)  
+- [SAM-Med2D](https://drive.google.com/file/d/1ARiB5RkSsWmAB_8mqWnwDF8ZKTtFwsjl/view?usp=drive_link)  
+
+</details>
+
+## 🔨 Usage
+### Quick Start for SAM-Med3D inference
+> **Note:**
+> Currently, labels are required to generate prompt points for inference.
+
+First, set up your environment with the following commands:
+```
+conda create --name sammed3d python=3.10 
+conda activate sammed3d
+pip install light-the-torch && ltt install torch
+pip install torchio opencv-python-headless matplotlib prefetch_generator monai edt medim
+```
+(if encounter OMP issue in Win, please refer to [link](https://github.com/uni-medical/SAM-Med3D/issues/103))
+
+Then, use [`medim_infer.py`](https://github.com/uni-medical/SAM-Med3D/blob/main/medim_infer.py) to test the inference:
+```
+python medim_infer.py
+```
+
+If you want to run inference on your own data, refer to [`medim_infer.py`](https://github.com/uni-medical/SAM-Med3D/blob/main/medim_infer.py) for more details. You can simply modify the paths in the script to use your own data. Here's the main logic:
+```
+  ''' 1. read and pre-process your input data '''
+  img_path = "./test_data/kidney_right/AMOS/imagesVal/amos_0013.nii.gz"
+  gt_path =  "./test_data/kidney_right/AMOS/labelsVal/amos_0013.nii.gz"
+  category_index = 3  # the index of your target category in the gt annotation
+  output_dir = "./test_data/kidney_right/AMOS/pred/"
+  roi_image, roi_label, meta_info = data_preprocess(img_path, gt_path, category_index=category_index)
+  
+  ''' 2. prepare the pre-trained model with local path or huggingface url '''
+  ckpt_path = "https://huggingface.co/blueyo0/SAM-Med3D/blob/main/sam_med3d_turbo.pth"
+  # or you can use the local path like: ckpt_path = "./ckpt/sam_med3d_turbo.pth"
+  model = medim.create_model("SAM-Med3D",
+                              pretrained=True,
+                              checkpoint_path=ckpt_path)
+  
+  ''' 3. infer with the pre-trained SAM-Med3D model '''
+  roi_pred = sam_model_infer(model, roi_image, roi_gt=roi_label)
+
+  ''' 4. post-process and save the result '''
+  output_path = osp.join(output_dir, osp.basename(img_path).replace(".nii.gz", "_pred.nii.gz"))
+  data_postprocess(roi_pred, meta_info, output_path, img_path)
+
+  print("result saved to", output_path)
+```
+
+
+### Training / Fine-tuning
+(we recommend fine-tuning with SAM-Med3D pre-trained weights from [link](https://github.com/uni-medical/SAM-Med3D#-checkpoint))
+
+To train the SAM-Med3D model on your own data, follow these steps:
+
+#### 0. **(Recommend) Prepare the Pre-trained Weights**
+
+> Note: You can easily get PyTorch SAM-Med3D model with pre-trained weights from huggingface use `MedIM`.
+> ```
+> ckpt_path = "https://huggingface.co/blueyo0/SAM-Med3D/blob/main/sam_med3d_turbo.pth"
+> model = medim.create_model("SAM-Med3D", pretrained=True, checkpoint_path=ckpt_path)
+> ```
+
+Download the checkpoint from [ckpt section](https://github.com/uni-medical/SAM-Med3D#-checkpoint) and move the pth file into `SAM_Med3D/ckpt/` (We recommand to use `SAM-Med3D-turbo.pth`).
+
+
+#### 1. Prepare Your Training Data (from nnU-Net-style dataset): 
+
+Ensure that your training data is organized according to the structure shown in the `data/medical_preprocessed` directories. The target file structures should be like the following:
+```
+data/medical_preprocessed
+      ├── adrenal
+      │ ├── ct_WORD
+      │ │ ├── imagesTr
+      │ │ │ ├── word_0025.nii.gz
+      │ │ │ ├── ...
+      │ │ ├── labelsTr
+      │ │ │ ├── word_0025.nii.gz
+      │ │ │ ├── ...
+      ├── ...
+```
+
+> If the original data are in the **nnU-Net style**, follow these steps:
+> 
+> For a nnU-Net style dataset, the original file structure should be:
+> ```
+> Task010_WORD
+>      ├── imagesTr
+>      │ ├── word_0025_0000.nii.gz
+>      │ ├── ...
+>      ├── labelsTr
+>      │ ├── word_0025.nii.gz
+>      │ ├── ...
+> ```
+> Then you should resample and convert the masks into binary. (You can use [script](https://github.com/uni-medical/SAM-Med3D/blob/b77585070b2f520ecd204b551a3f27715f5b3b43/utils/prepare_data_from_nnUNet.py) for nnU-Net folder)
+> ```
+> data/train
+>       ├── adrenal
+>       │ ├── ct_WORD
+>       │ │ ├── imagesTr
+>       │ │ │ ├── word_0025.nii.gz
+>       │ │ │ ├── ...
+>       │ │ ├── labelsTr
+>       │ │ │ ├── word_0025.nii.gz (binary label)
+>       │ │ │ ├── ...
+>       ├── liver
+>       │ ├── ct_WORD
+>       │ │ ├── imagesTr
+>       │ │ │ ├── word_0025.nii.gz
+>       │ │ │ ├── ...
+>       │ │ ├── labelsTr
+>       │ │ │ ├── word_0025.nii.gz (binary label)
+>       │ │ │ ├── ...
+>       ├── ...
+> ```
+
+Then, modify `img_datas` in `utils/data_paths.py` according to your own data.
+```
+img_datas = [
+"data/train/adrenal/ct_WORD",
+"data/train/liver/ct_WORD",
+...
+]
+```
+or
+```
+PROJ_DIR = <YOUR PROJ DIR>
+img_datas = glob(os.path.join(PROJ_DIR, "data", "train", "*", "*"))
+```
+
+
+#### 2. **Run the Training Script**: 
+Run `bash train.sh` to execute the following command in your terminal:
+
+```
+python train.py --multi_gpu --task_name ${tag}
+```
+This will start the training process of the SAM-Med3D model on your prepared data. If you use only one GPU, remove the `--multi_gpu` flag.
+
+The key options are listed below:
+
+- task_name: task name
+- checkpoint: pre-trained checkpoint
+- work_dir: results folder for log and ckpt
+- multi_gpu: use multiple GPU with DDP
+- gpu_ids: set gpu ids used for training
+- num_epochs: number of epoches
+- batch_size: batch size for training
+- lr: learning rate for training
+
+
+**Hint**: Use the `--checkpoint` to set the pre-trained weight path, the model will be trained from scratch if no ckpt in the path is found!
+
+### Evaluation & Inference
+Prepare your own dataset and refer to the samples in `data/validation` to replace them according to your specific scenario. 
+Then you can simply run `bash val.sh` to **quickly validate** SAM-Med3D on your data. Or you can use `bash infer.sh` to **generate full-volume results** for your application.
+Make sure the masks are processed into the one-hot format (have only two values: the main image (foreground) and the background). We highly recommend using the spacing of `1.5mm` for the best experience.
+
+```
+python validation.py --seed 2023\
+ -vp ./results/vis_sam_med3d \
+ -cp ./ckpt/sam_med3d_turbo.pth \
+ -tdp ./data/medical_preprocessed -nc 1 \
+ --save_name ./results/sam_med3d.py
+```
+
+- vp: visualization path, dir to save the final visualization files
+- cp: checkpoint path
+- tdp: test data path, where your data is placed
+- nc: number of clicks of prompt points
+- save_name: filename to save evaluation results 
+- (optional) skip_existing_pred: skip and not predict if output file is found existing
+
+**Sliding-window Inference (experimental)**: To extend the application scenario of SAM-Med3D and support more choices for full-volume inference. We provide the sliding-window mode here within `inference.py`. 
+```
+python inference.py --seed 2024\
+ -cp ./ckpt/sam_med3d_turbo.pth \
+ -tdp ./data/medical_preprocessed -nc 1 \
+ --output_dir ./results  --task_name test_amos_move \
+ --sliding_window --save_image_and_gt
+```
+- cp: checkpoint path
+- tdp: test data path, where your data is placed
+- output_dir&task_name: all your output will be saved to `<output_dir>/<task_name>`
+- (optional) sliding_window: enable the sliding-window mode. model will infer 27 patches with improved accuracy and slower responce.
+- (optional) save_image_and_gt: enable saving the full-volume image and ground-truth into `output_dir`, plz ensure your disk has enough free space when you turn on this
+
+For validation of SAM and SAM-Med2D on 3D volumetric data, you can refer to `scripts/val_sam.sh` and `scripts/val_med2d.sh` for details.
+
+Hint: We also provide a simple script `sum_result.py` to help summarize the results from files like `./results/sam_med3d.py`. 
+
+## 🗼 Method
+<div align="center">
+  <img src="assets/comparison.png">
+</div>
+<div align="center">
+  <img src="assets/architecture.png">
+</div>
+
+<!-- ## 🗓️ Ongoing 
+- [] Dataset release
+- [x] Train code release
+- [x] [Feature] Evaluation on 3D data with 2D models (slice-by-slice)
+- [x] Evaluation code release
+- [x] Pre-trained model release
+- [x] Paper release -->
+
+## 📬 Citation
+```
+@misc{wang2024sammed3dgeneralpurposesegmentationmodels,
+      title={SAM-Med3D: Towards General-purpose Segmentation Models for Volumetric Medical Images}, 
+      author={Haoyu Wang and Sizheng Guo and Jin Ye and Zhongying Deng and Junlong Cheng and Tianbin Li and Jianpin Chen and Yanzhou Su and Ziyan Huang and Yiqing Shen and Bin Fu and Shaoting Zhang and Junjun He and Yu Qiao},
+      year={2024},
+      eprint={2310.15161},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2310.15161}, 
+}
+```
+
+## 🎫 License
+This project is released under the [Apache 2.0 license](LICENSE). 
+
+## 💬 Discussion Group
+<p align="center"><img width="100" alt="image" src="assets/QRCode.jpg"></p> 
+(If the QRCode is expired, please contact the WeChat account: EugeneYonng or Small_dark8023，please note with "add sammed3d wechat"/请备注“sammed3d交流群”.)
+
+BTW, welcome to follow our [Zhihu official account](https://www.zhihu.com/people/gmai-38), we will share more information on medical imaging there.
+
+## 🙏 Acknowledgement
+- We thank all medical workers and dataset owners for making public datasets available to the community.
+- Thanks to the open-source of the following projects:
+  - [Segment Anything](https://github.com/facebookresearch/segment-anything) &#8194;
+  - [SAM-Med2D](https://github.com/OpenGVLab/SAM-Med2D/tree/main)
+
+## 👋 Hiring & Global Collaboration
+- **Hiring:** We are hiring researchers, engineers, and interns in General Vision Group, Shanghai AI Lab. If you are interested in Medical Foundation Models and General Medical AI, including designing benchmark datasets, general models, evaluation systems, and efficient tools, please contact us.
+- **Global Collaboration:** We're on a mission to redefine medical research, aiming for a more universally adaptable model. Our passionate team is delving into foundational healthcare models, promoting the development of the medical community. Collaborate with us to increase competitiveness, reduce risk, and expand markets.
+- **Contact:** Junjun He(hejunjun@pjlab.org.cn), Jin Ye(yejin@pjlab.org.cn), and Tianbin Li (litianbin@pjlab.org.cn).

requirements.txt

@@ -0,0 +1,8 @@
+numpy
+torch
+torchvision
+SimpleITK
+torchio
+opencv-python
+Pillow
+gradio

sample.py

@@ -0,0 +1,93 @@
+import SimpleITK as sitk
+import numpy as np
+import os
+
+def create_new_nifti_mask(output_nii_path, size=(256, 256, 128), mask_type="sphere"):
+    """
+    创建一个指定大小的新NIFTI掩码文件
+    
+    参数:
+        output_nii_path: 输出的伪掩码路径
+        size: 掩码的尺寸，默认为(256, 256, 128)
+        mask_type: 掩码类型，可选"sphere"、"cube"或"random"
+    """
+    print(f"创建尺寸为{size}的伪掩码")
+    
+    # 创建一个指定尺寸的零数组
+    mask_array = np.zeros(size, dtype=np.uint8)
+    
+    if mask_type == "sphere":
+        # 创建一个球形掩码
+        center = [s // 2 for s in size]
+        radius = min(size) // 4
+        
+        # 创建坐标网格
+        z, y, x = np.ogrid[:size[0], :size[1], :size[2]]
+        
+        # 计算到中心的距离
+        dist_from_center = np.sqrt(
+            (z - center[0])**2 + 
+            (y - center[1])**2 + 
+            (x - center[2])**2
+        )
+        
+        # 创建掩码
+        mask_array = (dist_from_center <= radius).astype(np.uint8)
+        
+    elif mask_type == "cube":
+        # 创建一个立方体掩码
+        center = [s // 2 for s in size]
+        half_width = min(size) // 6
+        
+        mask_array[
+            center[0]-half_width:center[0]+half_width,
+            center[1]-half_width:center[1]+half_width,
+            center[2]-half_width:center[2]+half_width
+        ] = 1
+        
+    elif mask_type == "random":
+        # 创建一个具有随机形状的掩码
+        np.random.seed(42)  # 设置随机数种子以便结果可复现
+        
+        # 创建几个随机椭球
+        for _ in range(3):
+            center = [np.random.randint(s//4, 3*s//4) for s in size]
+            radii = [s//10 + np.random.randint(s//8) for s in size]
+            
+            z, y, x = np.ogrid[:size[0], :size[1], :size[2]]
+            
+            dist_from_center = np.sqrt(
+                ((z - center[0])/radii[0])**2 + 
+                ((y - center[1])/radii[1])**2 + 
+                ((x - center[2])/radii[2])**2
+            )
+            
+            mask_array = np.logical_or(mask_array, dist_from_center <= 1)
+    
+    # 转换为SimpleITK图像
+    mask_image = sitk.GetImageFromArray(mask_array.astype(np.uint8))
+    
+    # 设置默认的元数据
+    mask_image.SetSpacing((1.0, 1.0, 1.0))  # 1mm 各向同性间距
+    mask_image.SetOrigin((0.0, 0.0, 0.0))    # 默认原点
+    direction = tuple([1.0 if i == j else 0.0 for i in range(3) for j in range(3)])  # 默认方向
+    mask_image.SetDirection(direction)
+    
+    # 确保输出目录存在
+    os.makedirs(os.path.dirname(output_nii_path), exist_ok=True)
+    
+    # 保存掩码
+    sitk.WriteImage(mask_image, output_nii_path)
+    print(f"伪掩码已保存至: {output_nii_path}")
+    
+    return mask_array
+
+# 使用相对路径
+input_path = "test_data/kidney_right/AMOS/imagesVal/test.nii.gz"
+output_path = "test_data/kidney_right/AMOS/labelsVal/test.nii.gz"
+
+# 确保父目录存在
+os.makedirs(os.path.dirname(output_path), exist_ok=True)
+
+# 创建新的掩码（尺寸可根据需要调整）
+create_new_nifti_mask(output_path, size=(128, 128, 128), mask_type="sphere")

scripts/val_default.sh

@@ -0,0 +1,5 @@
+python validation.py --seed 2023\
+ -vp ./results/vis_sam_med3d \
+ -cp ./ckpt/sam_med3d.pth \
+ -tdp ./data/validation_test1 -nc 10 \
+ --save_name ./results/sam_med3d.py

scripts/val_med2d.sh

@@ -0,0 +1,5 @@
+python validation.py --seed 2023\
+ -vp ./results/vis_sam_med2d \
+ -cp ./ckpt/sam_med2d.pth \
+ -tdp ./data/validation_test1 -nc 10 \
+ --image_size 256 -mt vit_b --dim 2 --save_name ./results/sam_med2d.py --ft2d 

scripts/val_sam.sh

@@ -0,0 +1,5 @@
+python validation.py --seed 2023\
+ -vp ./results/vis_sam_vit_b \
+ -cp ./ckpt/sam_vit_b.pth \
+ -tdp ./data/validation_test1 -nc 10 \
+ --image_size 1024 -mt vit_b --dim 2 --save_name ./results/sam_vit_b.py 

segment_anything/__init__.py

@@ -0,0 +1,11 @@
+from .build_sam3D import *
+from utils import *
+from .build_sam import (
+    build_sam,
+    build_sam_vit_h,
+    build_sam_vit_l,
+    build_sam_vit_b,
+    sam_model_registry,
+)
+from .predictor import SamPredictor
+from .automatic_mask_generator import SamAutomaticMaskGenerator

segment_anything/automatic_mask_generator.py

@@ -0,0 +1,372 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import numpy as np
+import torch
+from torchvision.ops.boxes import batched_nms, box_area  # type: ignore
+
+from typing import Any, Dict, List, Optional, Tuple
+
+from .modeling import Sam
+from .predictor import SamPredictor
+from .utils.amg import (
+    MaskData,
+    area_from_rle,
+    batch_iterator,
+    batched_mask_to_box,
+    box_xyxy_to_xywh,
+    build_all_layer_point_grids,
+    calculate_stability_score,
+    coco_encode_rle,
+    generate_crop_boxes,
+    is_box_near_crop_edge,
+    mask_to_rle_pytorch,
+    remove_small_regions,
+    rle_to_mask,
+    uncrop_boxes_xyxy,
+    uncrop_masks,
+    uncrop_points,
+)
+
+
+class SamAutomaticMaskGenerator:
+    def __init__(
+        self,
+        model: Sam,
+        points_per_side: Optional[int] = 32,
+        points_per_batch: int = 64,
+        pred_iou_thresh: float = 0.88,
+        stability_score_thresh: float = 0.95,
+        stability_score_offset: float = 1.0,
+        box_nms_thresh: float = 0.7,
+        crop_n_layers: int = 0,
+        crop_nms_thresh: float = 0.7,
+        crop_overlap_ratio: float = 512 / 1500,
+        crop_n_points_downscale_factor: int = 1,
+        point_grids: Optional[List[np.ndarray]] = None,
+        min_mask_region_area: int = 0,
+        output_mode: str = "binary_mask",
+    ) -> None:
+        """
+        Using a SAM model, generates masks for the entire image.
+        Generates a grid of point prompts over the image, then filters
+        low quality and duplicate masks. The default settings are chosen
+        for SAM with a ViT-H backbone.
+
+        Arguments:
+          model (Sam): The SAM model to use for mask prediction.
+          points_per_side (int or None): The number of points to be sampled
+            along one side of the image. The total number of points is
+            points_per_side**2. If None, 'point_grids' must provide explicit
+            point sampling.
+          points_per_batch (int): Sets the number of points run simultaneously
+            by the model. Higher numbers may be faster but use more GPU memory.
+          pred_iou_thresh (float): A filtering threshold in [0,1], using the
+            model's predicted mask quality.
+          stability_score_thresh (float): A filtering threshold in [0,1], using
+            the stability of the mask under changes to the cutoff used to binarize
+            the model's mask predictions.
+          stability_score_offset (float): The amount to shift the cutoff when
+            calculated the stability score.
+          box_nms_thresh (float): The box IoU cutoff used by non-maximal
+            suppression to filter duplicate masks.
+          crop_n_layers (int): If >0, mask prediction will be run again on
+            crops of the image. Sets the number of layers to run, where each
+            layer has 2**i_layer number of image crops.
+          crop_nms_thresh (float): The box IoU cutoff used by non-maximal
+            suppression to filter duplicate masks between different crops.
+          crop_overlap_ratio (float): Sets the degree to which crops overlap.
+            In the first crop layer, crops will overlap by this fraction of
+            the image length. Later layers with more crops scale down this overlap.
+          crop_n_points_downscale_factor (int): The number of points-per-side
+            sampled in layer n is scaled down by crop_n_points_downscale_factor**n.
+          point_grids (list(np.ndarray) or None): A list over explicit grids
+            of points used for sampling, normalized to [0,1]. The nth grid in the
+            list is used in the nth crop layer. Exclusive with points_per_side.
+          min_mask_region_area (int): If >0, postprocessing will be applied
+            to remove disconnected regions and holes in masks with area smaller
+            than min_mask_region_area. Requires opencv.
+          output_mode (str): The form masks are returned in. Can be 'binary_mask',
+            'uncompressed_rle', or 'coco_rle'. 'coco_rle' requires pycocotools.
+            For large resolutions, 'binary_mask' may consume large amounts of
+            memory.
+        """
+
+        assert (points_per_side is None) != (
+            point_grids is None
+        ), "Exactly one of points_per_side or point_grid must be provided."
+        if points_per_side is not None:
+            self.point_grids = build_all_layer_point_grids(
+                points_per_side,
+                crop_n_layers,
+                crop_n_points_downscale_factor,
+            )
+        elif point_grids is not None:
+            self.point_grids = point_grids
+        else:
+            raise ValueError("Can't have both points_per_side and point_grid be None.")
+
+        assert output_mode in [
+            "binary_mask",
+            "uncompressed_rle",
+            "coco_rle",
+        ], f"Unknown output_mode {output_mode}."
+        if output_mode == "coco_rle":
+            from pycocotools import mask as mask_utils  # type: ignore # noqa: F401
+
+        if min_mask_region_area > 0:
+            import cv2  # type: ignore # noqa: F401
+
+        self.predictor = SamPredictor(model)
+        self.points_per_batch = points_per_batch
+        self.pred_iou_thresh = pred_iou_thresh
+        self.stability_score_thresh = stability_score_thresh
+        self.stability_score_offset = stability_score_offset
+        self.box_nms_thresh = box_nms_thresh
+        self.crop_n_layers = crop_n_layers
+        self.crop_nms_thresh = crop_nms_thresh
+        self.crop_overlap_ratio = crop_overlap_ratio
+        self.crop_n_points_downscale_factor = crop_n_points_downscale_factor
+        self.min_mask_region_area = min_mask_region_area
+        self.output_mode = output_mode
+
+    @torch.no_grad()
+    def generate(self, image: np.ndarray) -> List[Dict[str, Any]]:
+        """
+        Generates masks for the given image.
+
+        Arguments:
+          image (np.ndarray): The image to generate masks for, in HWC uint8 format.
+
+        Returns:
+           list(dict(str, any)): A list over records for masks. Each record is
+             a dict containing the following keys:
+               segmentation (dict(str, any) or np.ndarray): The mask. If
+                 output_mode='binary_mask', is an array of shape HW. Otherwise,
+                 is a dictionary containing the RLE.
+               bbox (list(float)): The box around the mask, in XYWH format.
+               area (int): The area in pixels of the mask.
+               predicted_iou (float): The model's own prediction of the mask's
+                 quality. This is filtered by the pred_iou_thresh parameter.
+               point_coords (list(list(float))): The point coordinates input
+                 to the model to generate this mask.
+               stability_score (float): A measure of the mask's quality. This
+                 is filtered on using the stability_score_thresh parameter.
+               crop_box (list(float)): The crop of the image used to generate
+                 the mask, given in XYWH format.
+        """
+
+        # Generate masks
+        mask_data = self._generate_masks(image)
+
+        # Filter small disconnected regions and holes in masks
+        if self.min_mask_region_area > 0:
+            mask_data = self.postprocess_small_regions(
+                mask_data,
+                self.min_mask_region_area,
+                max(self.box_nms_thresh, self.crop_nms_thresh),
+            )
+
+        # Encode masks
+        if self.output_mode == "coco_rle":
+            mask_data["segmentations"] = [coco_encode_rle(rle) for rle in mask_data["rles"]]
+        elif self.output_mode == "binary_mask":
+            mask_data["segmentations"] = [rle_to_mask(rle) for rle in mask_data["rles"]]
+        else:
+            mask_data["segmentations"] = mask_data["rles"]
+
+        # Write mask records
+        curr_anns = []
+        for idx in range(len(mask_data["segmentations"])):
+            ann = {
+                "segmentation": mask_data["segmentations"][idx],
+                "area": area_from_rle(mask_data["rles"][idx]),
+                "bbox": box_xyxy_to_xywh(mask_data["boxes"][idx]).tolist(),
+                "predicted_iou": mask_data["iou_preds"][idx].item(),
+                "point_coords": [mask_data["points"][idx].tolist()],
+                "stability_score": mask_data["stability_score"][idx].item(),
+                "crop_box": box_xyxy_to_xywh(mask_data["crop_boxes"][idx]).tolist(),
+            }
+            curr_anns.append(ann)
+
+        return curr_anns
+
+    def _generate_masks(self, image: np.ndarray) -> MaskData:
+        orig_size = image.shape[:2]
+        crop_boxes, layer_idxs = generate_crop_boxes(
+            orig_size, self.crop_n_layers, self.crop_overlap_ratio
+        )
+
+        # Iterate over image crops
+        data = MaskData()
+        for crop_box, layer_idx in zip(crop_boxes, layer_idxs):
+            crop_data = self._process_crop(image, crop_box, layer_idx, orig_size)
+            data.cat(crop_data)
+
+        # Remove duplicate masks between crops
+        if len(crop_boxes) > 1:
+            # Prefer masks from smaller crops
+            scores = 1 / box_area(data["crop_boxes"])
+            scores = scores.to(data["boxes"].device)
+            keep_by_nms = batched_nms(
+                data["boxes"].float(),
+                scores,
+                torch.zeros_like(data["boxes"][:, 0]),  # categories
+                iou_threshold=self.crop_nms_thresh,
+            )
+            data.filter(keep_by_nms)
+
+        data.to_numpy()
+        return data
+
+    def _process_crop(
+        self,
+        image: np.ndarray,
+        crop_box: List[int],
+        crop_layer_idx: int,
+        orig_size: Tuple[int, ...],
+    ) -> MaskData:
+        # Crop the image and calculate embeddings
+        x0, y0, x1, y1 = crop_box
+        cropped_im = image[y0:y1, x0:x1, :]
+        cropped_im_size = cropped_im.shape[:2]
+        self.predictor.set_image(cropped_im)
+
+        # Get points for this crop
+        points_scale = np.array(cropped_im_size)[None, ::-1]
+        points_for_image = self.point_grids[crop_layer_idx] * points_scale
+
+        # Generate masks for this crop in batches
+        data = MaskData()
+        for (points,) in batch_iterator(self.points_per_batch, points_for_image):
+            batch_data = self._process_batch(points, cropped_im_size, crop_box, orig_size)
+            data.cat(batch_data)
+            del batch_data
+        self.predictor.reset_image()
+
+        # Remove duplicates within this crop.
+        keep_by_nms = batched_nms(
+            data["boxes"].float(),
+            data["iou_preds"],
+            torch.zeros_like(data["boxes"][:, 0]),  # categories
+            iou_threshold=self.box_nms_thresh,
+        )
+        data.filter(keep_by_nms)
+
+        # Return to the original image frame
+        data["boxes"] = uncrop_boxes_xyxy(data["boxes"], crop_box)
+        data["points"] = uncrop_points(data["points"], crop_box)
+        data["crop_boxes"] = torch.tensor([crop_box for _ in range(len(data["rles"]))])
+
+        return data
+
+    def _process_batch(
+        self,
+        points: np.ndarray,
+        im_size: Tuple[int, ...],
+        crop_box: List[int],
+        orig_size: Tuple[int, ...],
+    ) -> MaskData:
+        orig_h, orig_w = orig_size
+
+        # Run model on this batch
+        transformed_points = self.predictor.transform.apply_coords(points, im_size)
+        in_points = torch.as_tensor(transformed_points, device=self.predictor.device)
+        in_labels = torch.ones(in_points.shape[0], dtype=torch.int, device=in_points.device)
+        masks, iou_preds, _ = self.predictor.predict_torch(
+            in_points[:, None, :],
+            in_labels[:, None],
+            multimask_output=True,
+            return_logits=True,
+        )
+
+        # Serialize predictions and store in MaskData
+        data = MaskData(
+            masks=masks.flatten(0, 1),
+            iou_preds=iou_preds.flatten(0, 1),
+            points=torch.as_tensor(points.repeat(masks.shape[1], axis=0)),
+        )
+        del masks
+
+        # Filter by predicted IoU
+        if self.pred_iou_thresh > 0.0:
+            keep_mask = data["iou_preds"] > self.pred_iou_thresh
+            data.filter(keep_mask)
+
+        # Calculate stability score
+        data["stability_score"] = calculate_stability_score(
+            data["masks"], self.predictor.model.mask_threshold, self.stability_score_offset
+        )
+        if self.stability_score_thresh > 0.0:
+            keep_mask = data["stability_score"] >= self.stability_score_thresh
+            data.filter(keep_mask)
+
+        # Threshold masks and calculate boxes
+        data["masks"] = data["masks"] > self.predictor.model.mask_threshold
+        data["boxes"] = batched_mask_to_box(data["masks"])
+
+        # Filter boxes that touch crop boundaries
+        keep_mask = ~is_box_near_crop_edge(data["boxes"], crop_box, [0, 0, orig_w, orig_h])
+        if not torch.all(keep_mask):
+            data.filter(keep_mask)
+
+        # Compress to RLE
+        data["masks"] = uncrop_masks(data["masks"], crop_box, orig_h, orig_w)
+        data["rles"] = mask_to_rle_pytorch(data["masks"])
+        del data["masks"]
+
+        return data
+
+    @staticmethod
+    def postprocess_small_regions(
+        mask_data: MaskData, min_area: int, nms_thresh: float
+    ) -> MaskData:
+        """
+        Removes small disconnected regions and holes in masks, then reruns
+        box NMS to remove any new duplicates.
+
+        Edits mask_data in place.
+
+        Requires open-cv as a dependency.
+        """
+        if len(mask_data["rles"]) == 0:
+            return mask_data
+
+        # Filter small disconnected regions and holes
+        new_masks = []
+        scores = []
+        for rle in mask_data["rles"]:
+            mask = rle_to_mask(rle)
+
+            mask, changed = remove_small_regions(mask, min_area, mode="holes")
+            unchanged = not changed
+            mask, changed = remove_small_regions(mask, min_area, mode="islands")
+            unchanged = unchanged and not changed
+
+            new_masks.append(torch.as_tensor(mask).unsqueeze(0))
+            # Give score=0 to changed masks and score=1 to unchanged masks
+            # so NMS will prefer ones that didn't need postprocessing
+            scores.append(float(unchanged))
+
+        # Recalculate boxes and remove any new duplicates
+        masks = torch.cat(new_masks, dim=0)
+        boxes = batched_mask_to_box(masks)
+        keep_by_nms = batched_nms(
+            boxes.float(),
+            torch.as_tensor(scores),
+            torch.zeros_like(boxes[:, 0]),  # categories
+            iou_threshold=nms_thresh,
+        )
+
+        # Only recalculate RLEs for masks that have changed
+        for i_mask in keep_by_nms:
+            if scores[i_mask] == 0.0:
+                mask_torch = masks[i_mask].unsqueeze(0)
+                mask_data["rles"][i_mask] = mask_to_rle_pytorch(mask_torch)[0]
+                mask_data["boxes"][i_mask] = boxes[i_mask]  # update res directly
+        mask_data.filter(keep_by_nms)
+
+        return mask_data

segment_anything/build_sam.py

@@ -0,0 +1,161 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from functools import partial
+from .modeling import ImageEncoderViT, MaskDecoder, PromptEncoder, Sam, TwoWayTransformer
+from torch.nn import functional as F
+
+def build_sam_vit_h(args):
+    return _build_sam(
+        encoder_embed_dim=1280,
+        encoder_depth=32,
+        encoder_num_heads=16,
+        encoder_global_attn_indexes=[7, 15, 23, 31],
+        image_size=args.image_size,
+        checkpoint=args.sam_checkpoint,
+    )
+
+
+build_sam = build_sam_vit_h
+
+
+def build_sam_vit_l(args):
+    return _build_sam(
+        encoder_embed_dim=1024,
+        encoder_depth=24,
+        encoder_num_heads=16,
+        encoder_global_attn_indexes=[5, 11, 17, 23],
+        image_size=args.image_size,
+        checkpoint=args.sam_checkpoint,
+    )
+
+
+def build_sam_vit_b(args):
+    return _build_sam(
+        encoder_embed_dim=768,
+        encoder_depth=12,
+        encoder_num_heads=12,
+        encoder_global_attn_indexes=[2, 5, 8, 11],
+        image_size=args.image_size,
+        checkpoint=args.sam_checkpoint,
+
+    )
+
+
+sam_model_registry = {
+    "default": build_sam_vit_h,
+    "vit_h": build_sam_vit_h,
+    "vit_l": build_sam_vit_l,
+    "vit_b": build_sam_vit_b,
+}
+
+
+def _build_sam(
+    encoder_embed_dim,
+    encoder_depth,
+    encoder_num_heads,
+    encoder_global_attn_indexes,
+    image_size,
+    checkpoint,
+):
+    prompt_embed_dim = 256
+    image_size = image_size
+    vit_patch_size = 16
+    image_embedding_size = image_size // vit_patch_size
+    sam = Sam(
+        image_encoder=ImageEncoderViT(
+            depth=encoder_depth,
+            embed_dim=encoder_embed_dim,
+            img_size=image_size,
+            mlp_ratio=4,
+            norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
+            num_heads=encoder_num_heads,
+            patch_size=vit_patch_size,
+            qkv_bias=True,
+            use_rel_pos = True,
+            global_attn_indexes=encoder_global_attn_indexes,
+            window_size=14,
+            out_chans=prompt_embed_dim,
+        ),
+        prompt_encoder=PromptEncoder(
+            embed_dim=prompt_embed_dim,
+            image_embedding_size=(image_embedding_size, image_embedding_size),
+            input_image_size=(image_size, image_size),
+            mask_in_chans=16,
+        ),
+        mask_decoder=MaskDecoder(
+            num_multimask_outputs=3,
+            transformer=TwoWayTransformer(
+                depth=2,
+                embedding_dim=prompt_embed_dim,
+                mlp_dim=2048,
+                num_heads=8,
+            ),
+            transformer_dim=prompt_embed_dim,
+            iou_head_depth=3,
+            iou_head_hidden_dim=256,
+        ),
+        pixel_mean=[123.675, 116.28, 103.53],
+        pixel_std=[58.395, 57.12, 57.375],
+    )
+    sam.train()
+    if checkpoint is not None:
+        with open(checkpoint, "rb") as f:
+            state_dict = torch.load(f)
+        try:
+            if 'model' in state_dict.keys():
+                sam.load_state_dict(state_dict['model'])
+            else:
+                sam.load_state_dict(state_dict)
+        except:
+            print('*******interpolate')
+            new_state_dict = load_from(sam, state_dict, image_size, vit_patch_size)   
+            sam.load_state_dict(new_state_dict)
+        print(f"*******load {checkpoint}")
+        
+    return sam
+
+
+def load_from(sam, state_dicts, image_size, vit_patch_size):
+    sam_dict = sam.state_dict()
+    except_keys = ['mask_tokens', 'output_hypernetworks_mlps', 'iou_prediction_head']
+    new_state_dict = {k: v for k, v in state_dicts.items() if
+                      k in sam_dict.keys() and except_keys[0] not in k and except_keys[1] not in k and except_keys[2] not in k}
+    pos_embed = new_state_dict['image_encoder.pos_embed']
+    token_size = int(image_size // vit_patch_size)
+    if pos_embed.shape[1] != token_size:
+        # resize pos embedding, which may sacrifice the performance, but I have no better idea
+        pos_embed = pos_embed.permute(0, 3, 1, 2)  # [b, c, h, w]
+        pos_embed = F.interpolate(pos_embed, (token_size, token_size), mode='bilinear', align_corners=False)
+        pos_embed = pos_embed.permute(0, 2, 3, 1)  # [b, h, w, c]
+        new_state_dict['image_encoder.pos_embed'] = pos_embed
+        rel_pos_keys = [k for k in sam_dict.keys() if 'rel_pos' in k]
+
+        global_rel_pos_keys = [k for k in rel_pos_keys if 
+                                                        '2' in k or 
+                                                        '5' in k or 
+                                                        '7' in k or 
+                                                        '8' in k or 
+                                                        '11' in k or 
+                                                        '13' in k or
+                                                        '15' in k or 
+                                                        '23' in k or 
+                                                        '31' in k] 
+        # print(sam_dict)
+        for k in global_rel_pos_keys:
+            h_check, w_check = sam_dict[k].shape
+            rel_pos_params = new_state_dict[k]
+            h, w = rel_pos_params.shape
+            rel_pos_params = rel_pos_params.unsqueeze(0).unsqueeze(0)
+            if h != h_check or w != w_check:
+                rel_pos_params = F.interpolate(rel_pos_params, (h_check, w_check), mode='bilinear', align_corners=False)
+
+            new_state_dict[k] = rel_pos_params[0, 0, ...]
+
+    sam_dict.update(new_state_dict)
+    return sam_dict
+

segment_anything/build_sam3D.py

@@ -0,0 +1,161 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+
+from functools import partial
+
+from .modeling import ImageEncoderViT3D, MaskDecoder3D, PromptEncoder3D, Sam3D
+
+def build_sam3D_vit_h(checkpoint=None):
+    return _build_sam3D(
+        encoder_embed_dim=1280,
+        encoder_depth=32,
+        encoder_num_heads=16,
+        encoder_global_attn_indexes=[7, 15, 23, 31],
+        checkpoint=checkpoint,
+    )
+
+
+build_sam3D = build_sam3D_vit_h
+
+
+def build_sam3D_vit_l(checkpoint=None):
+    return _build_sam3D(
+        encoder_embed_dim=1024,
+        encoder_depth=24,
+        encoder_num_heads=16,
+        encoder_global_attn_indexes=[5, 11, 17, 23],
+        checkpoint=checkpoint,
+    )
+
+
+def build_sam3D_vit_b(checkpoint=None):
+    return _build_sam3D(
+        # encoder_embed_dim=768,
+        encoder_embed_dim=384,
+        encoder_depth=12,
+        encoder_num_heads=12,
+        encoder_global_attn_indexes=[2, 5, 8, 11],
+        checkpoint=checkpoint,
+    )
+
+def build_sam3D_vit_b_ori(checkpoint=None):
+    return _build_sam3D_ori(
+        encoder_embed_dim=768,
+        encoder_depth=12,
+        encoder_num_heads=12,
+        encoder_global_attn_indexes=[2, 5, 8, 11],
+        checkpoint=checkpoint,
+    )
+
+
+sam_model_registry3D = {
+    "default": build_sam3D_vit_h,
+    "vit_h": build_sam3D_vit_h,
+    "vit_l": build_sam3D_vit_l,
+    "vit_b": build_sam3D_vit_b,
+    "vit_b_ori": build_sam3D_vit_b_ori,
+}
+
+
+
+def _build_sam3D(
+    encoder_embed_dim,
+    encoder_depth,
+    encoder_num_heads,
+    encoder_global_attn_indexes,
+    checkpoint=None,
+):
+    prompt_embed_dim = 384
+    image_size = 256
+    vit_patch_size = 16
+    image_embedding_size = image_size // vit_patch_size
+    sam = Sam3D(
+        image_encoder=ImageEncoderViT3D(
+            depth=encoder_depth,
+            embed_dim=encoder_embed_dim,
+            img_size=image_size,
+            mlp_ratio=4,
+            norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
+            num_heads=encoder_num_heads,
+            patch_size=vit_patch_size,
+            qkv_bias=True,
+            use_rel_pos=True,
+            global_attn_indexes=encoder_global_attn_indexes,
+            window_size=14,
+            out_chans=prompt_embed_dim,
+        ),
+        prompt_encoder=PromptEncoder3D(
+            embed_dim=prompt_embed_dim,
+            image_embedding_size=(image_embedding_size, image_embedding_size, image_embedding_size),
+            input_image_size=(image_size, image_size, image_size),
+            mask_in_chans=16,
+        ),
+        mask_decoder=MaskDecoder3D(
+            num_multimask_outputs=3,
+            transformer_dim=prompt_embed_dim,
+            iou_head_depth=3,
+            iou_head_hidden_dim=256,
+        ),
+        pixel_mean=[123.675, 116.28, 103.53],
+        pixel_std=[58.395, 57.12, 57.375],
+    )
+    sam.eval()
+    if checkpoint is not None:
+        with open(checkpoint, "rb") as f:
+            state_dict = torch.load(f)
+        sam.load_state_dict(state_dict)
+    return sam
+
+
+def _build_sam3D_ori(
+    encoder_embed_dim,
+    encoder_depth,
+    encoder_num_heads,
+    encoder_global_attn_indexes,
+    checkpoint=None,
+):
+    prompt_embed_dim = 384
+    image_size = 128
+    vit_patch_size = 16
+    image_embedding_size = image_size // vit_patch_size
+    sam = Sam3D(
+        image_encoder=ImageEncoderViT3D(
+            depth=encoder_depth,
+            embed_dim=encoder_embed_dim,
+            img_size=image_size,
+            mlp_ratio=4,
+            norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
+            num_heads=encoder_num_heads,
+            patch_size=vit_patch_size,
+            qkv_bias=True,
+            use_rel_pos=True,
+            global_attn_indexes=encoder_global_attn_indexes,
+            window_size=14,
+            out_chans=prompt_embed_dim,
+        ),
+        prompt_encoder=PromptEncoder3D(
+            embed_dim=prompt_embed_dim,
+            image_embedding_size=(image_embedding_size, image_embedding_size, image_embedding_size),
+            input_image_size=(image_size, image_size, image_size),
+            mask_in_chans=16,
+        ),
+        mask_decoder=MaskDecoder3D(
+            num_multimask_outputs=3,
+            transformer_dim=prompt_embed_dim,
+            iou_head_depth=3,
+            iou_head_hidden_dim=256,
+        ),
+        pixel_mean=[123.675, 116.28, 103.53],
+        pixel_std=[58.395, 57.12, 57.375],
+    )
+    sam.eval()
+    if checkpoint is not None:
+        with open(checkpoint, "rb") as f:
+            state_dict = torch.load(f)
+        sam.load_state_dict(state_dict)
+    return sam

segment_anything/modeling/__init__.py

@@ -0,0 +1,10 @@
+from .sam3D import Sam3D
+from .image_encoder3D import ImageEncoderViT3D
+from .mask_decoder3D import MaskDecoder3D, TwoWayTransformer3D
+from .prompt_encoder3D import PromptEncoder3D
+
+from .sam_model import Sam
+from .image_encoder import ImageEncoderViT
+from .mask_decoder import MaskDecoder
+from .prompt_encoder import PromptEncoder
+from .transformer import TwoWayTransformer

segment_anything/modeling/common.py

@@ -0,0 +1,45 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+import torch.nn as nn
+
+from typing import Type
+
+
+class MLPBlock(nn.Module):
+    def __init__(
+        self,
+        embedding_dim: int,
+        mlp_dim: int,
+        act: Type[nn.Module] = nn.GELU,
+    ) -> None:
+        super().__init__()
+        self.lin1 = nn.Linear(embedding_dim, mlp_dim)
+        self.lin2 = nn.Linear(mlp_dim, embedding_dim)
+        self.act = act()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.lin2(self.act(self.lin1(x)))
+
+
+# From https://github.com/facebookresearch/detectron2/blob/main/detectron2/layers/batch_norm.py # noqa
+# Itself from https://github.com/facebookresearch/ConvNeXt/blob/d1fa8f6fef0a165b27399986cc2bdacc92777e40/models/convnext.py#L119  # noqa
+class LayerNorm2d(nn.Module):
+    def __init__(self, num_channels: int, eps: float = 1e-6) -> None:
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(num_channels))
+        self.bias = nn.Parameter(torch.zeros(num_channels))
+        self.eps = eps
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        u = x.mean(1, keepdim=True)
+        s = (x - u).pow(2).mean(1, keepdim=True)
+        x = (x - u) / torch.sqrt(s + self.eps)
+        y = self.weight[:, None, None] * x
+        # y = torch.mul(self.weight[:, None, None], x)
+        x = y + self.bias[:, None, None]
+        return x

segment_anything/modeling/image_encoder.py

@@ -0,0 +1,401 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from typing import Optional, Tuple, Type
+
+from .common import LayerNorm2d, MLPBlock
+
+
+# This class and its supporting functions below lightly adapted from the ViTDet backbone available at: https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/backbone/vit.py # noqa
+class ImageEncoderViT(nn.Module):
+    def __init__(
+        self,
+        img_size: int = 1024,
+        patch_size: int = 16,
+        in_chans: int = 3,
+        embed_dim: int = 768,
+        depth: int = 12,
+        num_heads: int = 12,
+        mlp_ratio: float = 4.0,
+        out_chans: int = 256,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        global_attn_indexes: Tuple[int, ...] = (),
+    ) -> None:
+        """
+        Args:
+            img_size (int): Input image size.
+            patch_size (int): Patch size.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+            depth (int): Depth of ViT.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Module): Normalization layer.
+            act_layer (nn.Module): Activation layer.
+            use_abs_pos (bool): If True, use absolute positional embeddings.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            window_size (int): Window size for window attention blocks.
+            global_attn_indexes (list): Indexes for blocks using global attention.
+        """
+        super().__init__()
+        self.img_size = img_size
+
+        self.patch_embed = PatchEmbed(
+            kernel_size=(patch_size, patch_size),
+            stride=(patch_size, patch_size),
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+        )
+
+        self.pos_embed: Optional[nn.Parameter] = None
+        if use_abs_pos:
+            # Initialize absolute positional embedding with pretrain image size.
+            self.pos_embed = nn.Parameter(
+                torch.zeros(1, img_size // patch_size, img_size // patch_size, embed_dim)
+            )
+
+
+        self.blocks = nn.ModuleList()
+        for i in range(depth):
+            block = Block(
+                dim=embed_dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                norm_layer=norm_layer,
+                act_layer=act_layer,
+                use_rel_pos=use_rel_pos,
+                rel_pos_zero_init=rel_pos_zero_init,
+                window_size=window_size if i not in global_attn_indexes else 0,
+                input_size=(img_size // patch_size, img_size // patch_size),
+            )
+            self.blocks.append(block)
+
+        self.neck = nn.Sequential(
+            nn.Conv2d(
+                embed_dim,
+                out_chans,
+                kernel_size=1,
+                bias=False,
+            ),
+            LayerNorm2d(out_chans),
+            nn.Conv2d(
+                out_chans,
+                out_chans,
+                kernel_size=3,
+                padding=1,
+                bias=False,
+            ),
+            LayerNorm2d(out_chans),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.patch_embed(x)
+        if self.pos_embed is not None:
+            x = x + self.pos_embed
+
+        for blk in self.blocks:
+            x = blk(x)
+
+
+        x = self.neck(x.permute(0, 3, 1, 2))
+
+        return x
+
+
+class Block(nn.Module):
+    """Transformer blocks with support of window attention and residual propagation blocks"""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int,
+        mlp_ratio: float = 4.0,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        input_size: Optional[Tuple[int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Module): Normalization layer.
+            act_layer (nn.Module): Activation layer.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            window_size (int): Window size for window attention blocks. If it equals 0, then
+                use global attention.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            use_rel_pos=use_rel_pos,
+            rel_pos_zero_init=rel_pos_zero_init,
+            input_size=input_size if window_size == 0 else (window_size, window_size),
+        )
+
+        self.norm2 = norm_layer(dim)
+        self.mlp = MLPBlock(embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer)
+
+        self.window_size = window_size
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        shortcut = x
+        x = self.norm1(x)
+        # Window partition
+        if self.window_size > 0:
+            H, W = x.shape[1], x.shape[2]
+            x, pad_hw = window_partition(x, self.window_size)
+
+        x = self.attn(x)
+        # Reverse window partition
+        if self.window_size > 0:
+            x = window_unpartition(x, self.window_size, pad_hw, (H, W))
+
+        x = shortcut + x
+        x = x + self.mlp(self.norm2(x))
+
+        return x
+
+
+class Attention(nn.Module):
+    """Multi-head Attention block with relative position embeddings."""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int = 8,
+        qkv_bias: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        input_size: Optional[Tuple[int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads.
+            qkv_bias (bool):  If True, add a learnable bias to query, key, value.
+            rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.proj = nn.Linear(dim, dim)
+
+        self.use_rel_pos = use_rel_pos
+        if self.use_rel_pos:
+            assert (
+                input_size is not None
+            ), "Input size must be provided if using relative positional encoding."
+            # initialize relative positional embeddings
+            self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
+            self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        B, H, W, _ = x.shape
+        # qkv with shape (3, B, nHead, H * W, C)
+        qkv = self.qkv(x).reshape(B, H * W, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
+        # q, k, v with shape (B * nHead, H * W, C)
+        q, k, v = qkv.reshape(3, B * self.num_heads, H * W, -1).unbind(0)
+
+        attn = (q * self.scale) @ k.transpose(-2, -1)
+
+        if self.use_rel_pos:
+            attn = add_decomposed_rel_pos(attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W))
+
+        attn = attn.softmax(dim=-1)
+        x = (attn @ v).view(B, self.num_heads, H, W, -1).permute(0, 2, 3, 1, 4).reshape(B, H, W, -1)
+        x = self.proj(x)
+
+        return x
+
+
+def window_partition(x: torch.Tensor, window_size: int) -> Tuple[torch.Tensor, Tuple[int, int]]:
+    """
+    Partition into non-overlapping windows with padding if needed.
+    Args:
+        x (tensor): input tokens with [B, H, W, C].
+        window_size (int): window size.
+
+    Returns:
+        windows: windows after partition with [B * num_windows, window_size, window_size, C].
+        (Hp, Wp): padded height and width before partition
+    """
+    B, H, W, C = x.shape
+
+    pad_h = (window_size - H % window_size) % window_size
+    pad_w = (window_size - W % window_size) % window_size
+    if pad_h > 0 or pad_w > 0:
+        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h))
+    Hp, Wp = H + pad_h, W + pad_w
+
+    x = x.view(B, Hp // window_size, window_size, Wp // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows, (Hp, Wp)
+
+
+def window_unpartition(
+    windows: torch.Tensor, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
+) -> torch.Tensor:
+    """
+    Window unpartition into original sequences and removing padding.
+    Args:
+        windows (tensor): input tokens with [B * num_windows, window_size, window_size, C].
+        window_size (int): window size.
+        pad_hw (Tuple): padded height and width (Hp, Wp).
+        hw (Tuple): original height and width (H, W) before padding.
+
+    Returns:
+        x: unpartitioned sequences with [B, H, W, C].
+    """
+    Hp, Wp = pad_hw
+    H, W = hw
+    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
+    x = windows.view(B, Hp // window_size, Wp // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, Hp, Wp, -1)
+
+    if Hp > H or Wp > W:
+        x = x[:, :H, :W, :].contiguous()
+    return x
+
+
+def get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor:
+    """
+    Get relative positional embeddings according to the relative positions of
+        query and key sizes.
+    Args:
+        q_size (int): size of query q.
+        k_size (int): size of key k.
+        rel_pos (Tensor): relative position embeddings (L, C).
+
+    Returns:
+        Extracted positional embeddings according to relative positions.
+    """
+    max_rel_dist = int(2 * max(q_size, k_size) - 1)
+    # Interpolate rel pos if needed.
+    if rel_pos.shape[0] != max_rel_dist:
+        # Interpolate rel pos.
+        rel_pos_resized = F.interpolate(
+            rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
+            size=max_rel_dist,
+            mode="linear",
+        )
+        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
+    else:
+        rel_pos_resized = rel_pos
+
+    # Scale the coords with short length if shapes for q and k are different.
+    q_coords = torch.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
+    k_coords = torch.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
+    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
+
+    return rel_pos_resized[relative_coords.long()]
+
+
+def add_decomposed_rel_pos(
+    attn: torch.Tensor,
+    q: torch.Tensor,
+    rel_pos_h: torch.Tensor,
+    rel_pos_w: torch.Tensor,
+    q_size: Tuple[int, int],
+    k_size: Tuple[int, int],
+) -> torch.Tensor:
+    """
+    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
+    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
+    Args:
+        attn (Tensor): attention map.
+        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
+        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
+        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
+        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
+        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).
+
+    Returns:
+        attn (Tensor): attention map with added relative positional embeddings.
+    """
+    q_h, q_w = q_size
+    k_h, k_w = k_size
+    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
+    Rw = get_rel_pos(q_w, k_w, rel_pos_w)
+
+    B, _, dim = q.shape
+    r_q = q.reshape(B, q_h, q_w, dim)
+
+    # r_q = r_q.to(torch.float) #todo   opt_level="O2" 模式下需要注释
+    r_q = r_q.to(Rh.dtype) #todo
+    
+    rel_h = torch.einsum("bhwc,hkc->bhwk", r_q, Rh)
+    rel_w = torch.einsum("bhwc,wkc->bhwk", r_q, Rw)
+
+    attn = (
+        attn.view(B, q_h, q_w, k_h, k_w) + rel_h[:, :, :, :, None] + rel_w[:, :, :, None, :]
+    ).view(B, q_h * q_w, k_h * k_w)
+
+    return attn
+
+
+class PatchEmbed(nn.Module):
+    """
+    Image to Patch Embedding.
+    """
+
+    def __init__(
+        self,
+        kernel_size: Tuple[int, int] = (16, 16),
+        stride: Tuple[int, int] = (16, 16),
+        padding: Tuple[int, int] = (0, 0),
+        in_chans: int = 3,
+        embed_dim: int = 768,
+    ) -> None:
+        """
+        Args:
+            kernel_size (Tuple): kernel size of the projection layer.
+            stride (Tuple): stride of the projection layer.
+            padding (Tuple): padding size of the projection layer.
+            in_chans (int): Number of input image channels.
+            embed_dim (int):  embed_dim (int): Patch embedding dimension.
+        """
+        super().__init__()
+
+        self.proj = nn.Conv2d(
+            in_chans, embed_dim, kernel_size=kernel_size, stride=stride, padding=padding
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.proj(x)
+        # B C H W -> B H W C
+        x = x.permute(0, 2, 3, 1)
+        return x

segment_anything/modeling/image_encoder3D.py

@@ -0,0 +1,442 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from typing import Optional, Tuple, Type
+
+
+class MLPBlock(nn.Module):
+    def __init__(
+        self,
+        embedding_dim: int,
+        mlp_dim: int,
+        act: Type[nn.Module] = nn.GELU,
+    ) -> None:
+        super().__init__()
+        self.lin1 = nn.Linear(embedding_dim, mlp_dim)
+        self.lin2 = nn.Linear(mlp_dim, embedding_dim)
+        self.act = act()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.lin2(self.act(self.lin1(x)))
+    
+class LayerNorm3d(nn.Module):
+    def __init__(self, num_channels: int, eps: float = 1e-6) -> None:
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(num_channels))
+        self.bias = nn.Parameter(torch.zeros(num_channels))
+        self.eps = eps
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        u = x.mean(1, keepdim=True)
+        s = (x - u).pow(2).mean(1, keepdim=True)
+        x = (x - u) / torch.sqrt(s + self.eps)
+        x = self.weight[:, None, None, None] * x + self.bias[:, None, None, None]
+        return x
+
+
+# This class and its supporting functions below lightly adapted from the ViTDet backbone available at: https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/backbone/vit.py # noqa
+class ImageEncoderViT3D(nn.Module):
+    def __init__(
+        self,
+        img_size: int = 256,
+        patch_size: int = 16,
+        in_chans: int = 1,
+        embed_dim: int = 768,
+        depth: int = 12,
+        num_heads: int = 12,
+        mlp_ratio: float = 4.0,
+        out_chans: int = 256,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        global_attn_indexes: Tuple[int, ...] = (),
+    ) -> None:
+        """
+        Args:
+            img_size (int): Input image size.
+            patch_size (int): Patch size.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+            depth (int): Depth of ViT.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Module): Normalization layer.
+            act_layer (nn.Module): Activation layer.
+            use_abs_pos (bool): If True, use absolute positional embeddings.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            window_size (int): Window size for window attention blocks.
+            global_attn_indexes (list): Indexes for blocks using global attention.
+        """
+        super().__init__()
+        self.img_size = img_size
+
+        self.patch_embed = PatchEmbed3D(
+            kernel_size=(patch_size, patch_size, patch_size),
+            stride=(patch_size, patch_size, patch_size),
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+        )
+
+        self.pos_embed: Optional[nn.Parameter] = None
+        if use_abs_pos:
+            # Initialize absolute positional embedding with pretrain image size.
+            self.pos_embed = nn.Parameter(
+                torch.zeros(1, img_size // patch_size, img_size // patch_size, img_size // patch_size, embed_dim)
+            )
+
+        self.blocks = nn.ModuleList()
+        for i in range(depth):
+            block = Block3D(
+                dim=embed_dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                norm_layer=norm_layer,
+                act_layer=act_layer,
+                use_rel_pos=use_rel_pos,
+                rel_pos_zero_init=rel_pos_zero_init,
+                window_size=window_size if i not in global_attn_indexes else 0,
+                input_size=(img_size // patch_size, img_size // patch_size, img_size // patch_size),
+            )
+            self.blocks.append(block)
+
+        self.neck = nn.Sequential(
+            nn.Conv3d(
+                embed_dim,
+                out_chans,
+                kernel_size=1,
+                bias=False,
+            ),
+            # nn.LayerNorm(out_chans),
+            LayerNorm3d(out_chans),
+            nn.Conv3d(
+                out_chans,
+                out_chans,
+                kernel_size=3,
+                padding=1,
+                bias=False,
+            ),
+            LayerNorm3d(out_chans),
+            # nn.LayerNorm(out_chans),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        # input_size = [1,1,256,256,256]
+        # import IPython; IPython.embed()
+        x = self.patch_embed(x)
+        # x = [1,16,16,16,768]
+        # import pdb; pdb.set_trace()
+        if self.pos_embed is not None:
+            x = x + self.pos_embed
+
+        for blk in self.blocks:
+            x = blk(x)
+        # x = [1,16,16,16,768]
+        x = self.neck(x.permute(0, 4, 1, 2, 3))
+
+        # output_size = [1,256,16,16,16]
+        return x
+
+
+class Block3D(nn.Module):
+    """Transformer blocks with support of window attention and residual propagation blocks"""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int,
+        mlp_ratio: float = 4.0,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        input_size: Optional[Tuple[int, int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Module): Normalization layer.
+            act_layer (nn.Module): Activation layer.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            window_size (int): Window size for window attention blocks. If it equals 0, then
+                use global attention.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            use_rel_pos=use_rel_pos,
+            rel_pos_zero_init=rel_pos_zero_init,
+            input_size=input_size if window_size == 0 else (window_size, window_size, window_size),
+        )
+
+        self.norm2 = norm_layer(dim)
+        self.mlp = MLPBlock(embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer)
+
+        self.window_size = window_size
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        shortcut = x
+        x = self.norm1(x)
+        # Window partition
+        if self.window_size > 0:
+            D, H, W = x.shape[1], x.shape[2], x.shape[3]
+            x, pad_dhw = window_partition3D(x, self.window_size)
+
+        x = self.attn(x)
+        # Reverse window partition
+        if self.window_size > 0:
+            x = window_unpartition3D(x, self.window_size, pad_dhw, (D, H, W))
+
+        x = shortcut + x
+        x = x + self.mlp(self.norm2(x))
+
+        return x
+
+
+class Attention(nn.Module):
+    """Multi-head Attention block with relative position embeddings."""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int = 8,
+        qkv_bias: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        input_size: Optional[Tuple[int, int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads.
+            qkv_bias (bool):  If True, add a learnable bias to query, key, value.
+            rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.proj = nn.Linear(dim, dim)
+
+        self.use_rel_pos = use_rel_pos
+        if self.use_rel_pos:
+            assert (
+                input_size is not None
+            ), "Input size must be provided if using relative positional encoding."
+            # initialize relative positional embeddings
+            self.rel_pos_d = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
+            self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))
+            self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[2] - 1, head_dim))
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        B, D, H, W, _ = x.shape
+        # qkv with shape (3, B, nHead, H * W, C)
+        qkv = self.qkv(x).reshape(B, D * H * W, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
+        # q, k, v with shape (B * nHead, H * W, C)
+        q, k, v = qkv.reshape(3, B * self.num_heads, D * H * W, -1).unbind(0)
+
+        attn = (q * self.scale) @ k.transpose(-2, -1)
+
+        if self.use_rel_pos:
+            attn = add_decomposed_rel_pos(attn, q, self.rel_pos_d, self.rel_pos_h, self.rel_pos_w, (D, H, W), (D, H, W))
+
+        attn = attn.softmax(dim=-1)
+        x = (attn @ v).view(B, self.num_heads, D, H, W, -1).permute(0, 2, 3, 4, 1, 5).reshape(B, D, H, W, -1)
+        x = self.proj(x)
+
+        return x
+
+
+def window_partition3D(x: torch.Tensor, window_size: int) -> Tuple[torch.Tensor, Tuple[int, int, int]]:
+    """
+    Partition into non-overlapping windows with padding if needed.
+    Args:
+        x (tensor): input tokens with [B, H, W, C].
+        window_size (int): window size.
+
+    Returns:
+        windows: windows after partition with [B * num_windows, window_size, window_size, C].
+        (Hp, Wp): padded height and width before partition
+    """
+    B, D, H, W, C = x.shape
+
+    pad_d = (window_size - D % window_size) % window_size
+    pad_h = (window_size - H % window_size) % window_size
+    pad_w = (window_size - W % window_size) % window_size
+    
+    if pad_h > 0 or pad_w > 0 or pad_d > 0:
+        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h, 0, pad_d))
+    Hp, Wp, Dp = H + pad_h, W + pad_w, D + pad_d
+
+    x = x.view(B, Dp // window_size, window_size, Hp // window_size, window_size, Wp // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 5, 2, 4, 6, 7).contiguous().view(-1, window_size, window_size, window_size, C)
+    return windows, (Dp, Hp, Wp)
+
+
+def window_unpartition3D(
+    windows: torch.Tensor, window_size: int, pad_dhw: Tuple[int, int, int], dhw: Tuple[int, int, int]
+) -> torch.Tensor:
+    """
+    Window unpartition into original sequences and removing padding.
+    Args:
+        windows (tensor): input tokens with [B * num_windows, window_size, window_size, C].
+        window_size (int): window size.
+        pad_hw (Tuple): padded height and width (Hp, Wp).
+        hw (Tuple): original height and width (H, W) before padding.
+
+    Returns:
+        x: unpartitioned sequences with [B, H, W, C].
+    """
+    Dp, Hp, Wp = pad_dhw
+    D, H, W = dhw
+    B = windows.shape[0] // (Dp * Hp * Wp // window_size // window_size // window_size)
+    x = windows.view(B, Dp // window_size, Hp // window_size, Wp // window_size, window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 4, 2, 5, 3, 6, 7).contiguous().view(B, Dp, Hp, Wp, -1)
+
+    if Hp > H or Wp > W or Dp > D:
+        x = x[:, :D, :H, :W, :].contiguous()
+    return x
+
+
+def get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor:
+    """
+    Get relative positional embeddings according to the relative positions of
+        query and key sizes.
+    Args:
+        q_size (int): size of query q.
+        k_size (int): size of key k.
+        rel_pos (Tensor): relative position embeddings (L, C).
+
+    Returns:
+        Extracted positional embeddings according to relative positions.
+    """
+    max_rel_dist = int(2 * max(q_size, k_size) - 1)
+    # Interpolate rel pos if needed.
+    if rel_pos.shape[0] != max_rel_dist:
+        # Interpolate rel pos.
+        rel_pos_resized = F.interpolate(
+            rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
+            size=max_rel_dist,
+            mode="linear",
+        )
+        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
+    else:
+        rel_pos_resized = rel_pos
+
+    # Scale the coords with short length if shapes for q and k are different.
+    q_coords = torch.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
+    k_coords = torch.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
+    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
+
+    return rel_pos_resized[relative_coords.long()]
+
+
+def add_decomposed_rel_pos(
+    attn: torch.Tensor,
+    q: torch.Tensor,
+    rel_pos_d: torch.Tensor,
+    rel_pos_h: torch.Tensor,
+    rel_pos_w: torch.Tensor,
+    q_size: Tuple[int, int, int],
+    k_size: Tuple[int, int, int],
+) -> torch.Tensor:
+    """
+    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
+    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
+    Args:
+        attn (Tensor): attention map.
+        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
+        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
+        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
+        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
+        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).
+
+    Returns:
+        attn (Tensor): attention map with added relative positional embeddings.
+    """
+    q_d, q_h, q_w = q_size
+    k_d, k_h, k_w = k_size
+
+    Rd = get_rel_pos(q_d, k_d, rel_pos_d)
+    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
+    Rw = get_rel_pos(q_w, k_w, rel_pos_w)
+    
+    B, _, dim = q.shape
+    r_q = q.reshape(B, q_d, q_h, q_w, dim)
+
+    rel_d = torch.einsum("bdhwc,dkc->bdhwk", r_q, Rd)
+    rel_h = torch.einsum("bdhwc,hkc->bdhwk", r_q, Rh)
+    rel_w = torch.einsum("bdhwc,wkc->bdhwk", r_q, Rw)
+    
+
+    
+    attn = (
+        attn.view(B, q_d, q_h, q_w, k_d, k_h, k_w) + rel_d[:, :, :, :, None, None] + rel_h[:, :, :, None, :, None] + rel_w[:, :, :,None,None, :]
+    ).view(B, q_d * q_h * q_w, k_d * k_h * k_w)
+
+    return attn
+
+
+class PatchEmbed3D(nn.Module):
+    """
+    Image to Patch Embedding.
+    """
+
+    def __init__(
+        self,
+        kernel_size: Tuple[int, int] = (16, 16, 16),
+        stride: Tuple[int, int] = (16, 16, 16),
+        padding: Tuple[int, int] = (0, 0, 0),
+        in_chans: int = 1,
+        embed_dim: int = 768,
+    ) -> None:
+        """
+        Args:
+            kernel_size (Tuple): kernel size of the projection layer.
+            stride (Tuple): stride of the projection layer.
+            padding (Tuple): padding size of the projection layer.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+        """
+        super().__init__()
+
+        self.proj = nn.Conv3d(
+            in_chans, embed_dim, kernel_size=kernel_size, stride=stride, padding=padding
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.proj(x)
+        # B C X Y Z -> B X Y Z C
+        x = x.permute(0, 2, 3, 4, 1)
+        return x
+
+

segment_anything/modeling/mask_decoder.py

@@ -0,0 +1,186 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from typing import List, Tuple, Type
+
+from .common import LayerNorm2d
+
+
+class MaskDecoder(nn.Module):
+    def __init__(
+        self,
+        *,
+        transformer_dim: int,
+        transformer: nn.Module,
+        num_multimask_outputs: int = 3,
+        activation: Type[nn.Module] = nn.GELU,
+        iou_head_depth: int = 3,
+        iou_head_hidden_dim: int = 256,
+    ) -> None:
+        """
+        Predicts masks given an image and prompt embeddings, using a
+        transformer architecture.
+
+        Arguments:
+          transformer_dim (int): the channel dimension of the transformer
+          transformer (nn.Module): the transformer used to predict masks
+          num_multimask_outputs (int): the number of masks to predict
+            when disambiguating masks
+          activation (nn.Module): the type of activation to use when
+            upscaling masks
+          iou_head_depth (int): the depth of the MLP used to predict
+            mask quality
+          iou_head_hidden_dim (int): the hidden dimension of the MLP
+            used to predict mask quality
+        """
+        super().__init__()
+        self.transformer_dim = transformer_dim
+        self.transformer = transformer
+
+        self.num_multimask_outputs = num_multimask_outputs
+
+        self.iou_token = nn.Embedding(1, transformer_dim)
+        self.num_mask_tokens = num_multimask_outputs + 1
+        self.mask_tokens = nn.Embedding(self.num_mask_tokens, transformer_dim)
+
+        self.output_upscaling = nn.Sequential(
+            nn.ConvTranspose2d(transformer_dim, transformer_dim // 4, kernel_size=2, stride=2),
+            LayerNorm2d(transformer_dim // 4),
+            activation(),
+            nn.ConvTranspose2d(transformer_dim // 4, transformer_dim // 8, kernel_size=2, stride=2),
+            activation(),
+        )
+        self.output_hypernetworks_mlps = nn.ModuleList(
+            [
+                MLP(transformer_dim, transformer_dim, transformer_dim // 8, 3)
+                for i in range(self.num_mask_tokens)
+            ]
+        )
+
+        self.iou_prediction_head = MLP(
+            transformer_dim, iou_head_hidden_dim, self.num_mask_tokens, iou_head_depth
+        )  #256 256 4 3
+
+    def forward(
+        self,
+        image_embeddings: torch.Tensor,   #[B, 256, 64, 64]
+        image_pe: torch.Tensor,           #[1, 256, 64, 64]
+        sparse_prompt_embeddings: torch.Tensor, #[B, 3, 256]
+        dense_prompt_embeddings: torch.Tensor,  #[B, 256, 64, 64]
+        multimask_output: bool,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Predict masks given image and prompt embeddings.
+
+        Arguments:
+          image_embeddings (torch.Tensor): the embeddings from the image encoder
+          image_pe (torch.Tensor): positional encoding with the shape of image_embeddings
+          sparse_prompt_embeddings (torch.Tensor): the embeddings of the points and boxes
+          dense_prompt_embeddings (torch.Tensor): the embeddings of the mask inputs
+          multimask_output (bool): Whether to return multiple masks or a single
+            mask.
+
+        Returns:
+          torch.Tensor: batched predicted masks
+          torch.Tensor: batched predictions of mask quality
+        """
+
+        masks, iou_pred = self.predict_masks(
+            image_embeddings=image_embeddings,
+            image_pe=image_pe,
+            sparse_prompt_embeddings=sparse_prompt_embeddings,
+            dense_prompt_embeddings=dense_prompt_embeddings,
+        )
+
+        # Select the correct mask or masks for output
+        if multimask_output:
+            mask_slice = slice(1, None)
+        else:
+            mask_slice = slice(0, 1)
+        masks = masks[:, mask_slice, :, :]
+        iou_pred = iou_pred[:, mask_slice]
+
+        # Prepare output
+        return masks, iou_pred
+
+    def predict_masks(
+        self,
+        image_embeddings: torch.Tensor,
+        image_pe: torch.Tensor,
+        sparse_prompt_embeddings: torch.Tensor,
+        dense_prompt_embeddings: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Predicts masks. See 'forward' for more details."""
+        # Concatenate output tokens
+
+        output_tokens = torch.cat([self.iou_token.weight, self.mask_tokens.weight], dim=0)  #iou_token:[1,256]  mask_tokens:[4,256]
+        output_tokens = output_tokens.unsqueeze(0).expand(sparse_prompt_embeddings.size(0), -1, -1)
+        tokens = torch.cat((output_tokens, sparse_prompt_embeddings), dim=1)
+
+        # Expand per-image data in batch direction to be per-mask
+        # src = torch.repeat_interleave(image_embeddings, tokens.shape[0], dim=0)
+        src = image_embeddings
+        src = src + dense_prompt_embeddings
+        pos_src = torch.repeat_interleave(image_pe, tokens.shape[0], dim=0)
+        b, c, h, w = src.shape
+
+        # Run the transformer
+        hs, src = self.transformer(src, pos_src, tokens)
+        iou_token_out = hs[:, 0, :]
+        mask_tokens_out = hs[:, 1 : (1 + self.num_mask_tokens), :]
+
+        # Upscale mask embeddings and predict masks using the mask tokens
+        src = src.transpose(1, 2).view(b, c, h, w)
+        upscaled_embedding = self.output_upscaling(src)
+        hyper_in_list: List[torch.Tensor] = []
+        for i in range(self.num_mask_tokens):
+            hyper_in_list.append(self.output_hypernetworks_mlps[i](mask_tokens_out[:, i, :]))
+        hyper_in = torch.stack(hyper_in_list, dim=1)  #[1,4,32]
+
+        b, c, h, w = upscaled_embedding.shape  #[1, 32, 256, 256]
+        masks = (hyper_in @ upscaled_embedding.view(b, c, h * w)).view(b, -1, h, w)
+
+        # Generate mask quality predictions
+        iou_pred = self.iou_prediction_head(iou_token_out)
+
+        return masks, iou_pred
+
+
+# Lightly adapted from
+# https://github.com/facebookresearch/MaskFormer/blob/main/mask_former/modeling/transformer/transformer_predictor.py # noqa
+class MLP(nn.Module):
+    def __init__(
+        self,
+        input_dim: int,
+        hidden_dim: int,
+        output_dim: int,
+        num_layers: int,
+        sigmoid_output: bool = False,
+    ) -> None:
+        super().__init__()
+        self.num_layers = num_layers
+        h = [hidden_dim] * (num_layers - 1)
+        self.layers = nn.ModuleList(
+            nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim])
+        )
+        self.sigmoid_output = sigmoid_output
+        self.relu = nn.ReLU(inplace=False)
+    def forward(self, x):
+        for i, layer in enumerate(self.layers):
+            # x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x)
+            # x = self.relu(layer(x)) if i < self.num_layers - 1 else layer(x) #源码
+            if i < self.num_layers - 1:
+                x = F.relu(layer(x))
+            else:
+                x = layer(x)
+
+        if self.sigmoid_output:
+            x = F.sigmoid(x)
+        return x

segment_anything/modeling/mask_decoder3D.py

@@ -0,0 +1,458 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from typing import List, Tuple, Type
+# from .transformer import TwoWayTransformer
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torch import Tensor, nn
+
+import math
+from typing import Tuple, Type
+
+
+class MLPBlock3D(nn.Module):
+    def __init__(
+        self,
+        embedding_dim: int,
+        mlp_dim: int,
+        act: Type[nn.Module] = nn.GELU,
+    ) -> None:
+        super().__init__()
+        self.lin1 = nn.Linear(embedding_dim, mlp_dim)
+        self.lin2 = nn.Linear(mlp_dim, embedding_dim)
+        self.act = act()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.lin2(self.act(self.lin1(x)))
+
+class TwoWayTransformer3D(nn.Module):
+    def __init__(
+        self,
+        depth: int,
+        embedding_dim: int,
+        num_heads: int,
+        mlp_dim: int,
+        activation: Type[nn.Module] = nn.ReLU,
+        attention_downsample_rate: int = 2,
+    ) -> None:
+        """
+        A transformer decoder that attends to an input image using
+        queries whose positional embedding is supplied.
+
+        Args:
+          depth (int): number of layers in the transformer
+          embedding_dim (int): the channel dimension for the input embeddings
+          num_heads (int): the number of heads for multihead attention. Must
+            divide embedding_dim
+          mlp_dim (int): the channel dimension internal to the MLP block
+          activation (nn.Module): the activation to use in the MLP block
+        """
+        super().__init__()
+        self.depth = depth
+        self.embedding_dim = embedding_dim
+        self.num_heads = num_heads
+        self.mlp_dim = mlp_dim
+        self.layers = nn.ModuleList()
+
+        for i in range(depth):
+            self.layers.append(
+                TwoWayAttentionBlock3D(
+                    embedding_dim=embedding_dim,
+                    num_heads=num_heads,
+                    mlp_dim=mlp_dim,
+                    activation=activation,
+                    attention_downsample_rate=attention_downsample_rate,
+                    skip_first_layer_pe=(i == 0),
+                )
+            )
+
+        self.final_attn_token_to_image = Attention(
+            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
+        )
+        self.norm_final_attn = nn.LayerNorm(embedding_dim)
+
+    def forward(
+        self,
+        image_embedding: Tensor,
+        image_pe: Tensor,
+        point_embedding: Tensor,
+    ) -> Tuple[Tensor, Tensor]:
+        """
+        Args:
+          image_embedding (torch.Tensor): image to attend to. Should be shape
+            B x embedding_dim x h x w for any h and w.
+          image_pe (torch.Tensor): the positional encoding to add to the image. Must
+            have the same shape as image_embedding.
+          point_embedding (torch.Tensor): the embedding to add to the query points.
+            Must have shape B x N_points x embedding_dim for any N_points.
+
+        Returns:
+          torch.Tensor: the processed point_embedding
+          torch.Tensor: the processed image_embedding
+        """
+        # BxCxHxW -> BxHWxC == B x N_image_tokens x C
+        bs, c, x, y, z = image_embedding.shape
+        image_embedding = image_embedding.flatten(2).permute(0, 2, 1)
+        image_pe = image_pe.flatten(2).permute(0, 2, 1)
+
+        # Prepare queries
+        queries = point_embedding
+        keys = image_embedding
+
+        # Apply transformer blocks and final layernorm
+        for layer in self.layers:
+            queries, keys = layer(
+                queries=queries,
+                keys=keys,
+                query_pe=point_embedding,
+                key_pe=image_pe,
+            )
+
+        # Apply the final attention layer from the points to the image
+        q = queries + point_embedding
+        k = keys + image_pe
+        attn_out = self.final_attn_token_to_image(q=q, k=k, v=keys)
+        queries = queries + attn_out
+        queries = self.norm_final_attn(queries)
+
+        return queries, keys
+
+
+class TwoWayAttentionBlock3D(nn.Module):
+    def __init__(
+        self,
+        embedding_dim: int,
+        num_heads: int,
+        mlp_dim: int = 2048,
+        activation: Type[nn.Module] = nn.ReLU,
+        attention_downsample_rate: int = 2,
+        skip_first_layer_pe: bool = False,
+    ) -> None:
+        """
+        A transformer block with four layers: (1) self-attention of sparse
+        inputs, (2) cross attention of sparse inputs to dense inputs, (3) mlp
+        block on sparse inputs, and (4) cross attention of dense inputs to sparse
+        inputs.
+
+        Arguments:
+          embedding_dim (int): the channel dimension of the embeddings
+          num_heads (int): the number of heads in the attention layers
+          mlp_dim (int): the hidden dimension of the mlp block
+          activation (nn.Module): the activation of the mlp block
+          skip_first_layer_pe (bool): skip the PE on the first layer
+        """
+        super().__init__()
+        self.self_attn = Attention(embedding_dim, num_heads)
+        self.norm1 = nn.LayerNorm(embedding_dim)
+
+        self.cross_attn_token_to_image = Attention(
+            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
+        )
+        self.norm2 = nn.LayerNorm(embedding_dim)
+
+        self.mlp = MLPBlock3D(embedding_dim, mlp_dim, activation)
+        self.norm3 = nn.LayerNorm(embedding_dim)
+
+        self.norm4 = nn.LayerNorm(embedding_dim)
+        self.cross_attn_image_to_token = Attention(
+            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
+        )
+
+        self.skip_first_layer_pe = skip_first_layer_pe
+
+    def forward(
+        self, queries: Tensor, keys: Tensor, query_pe: Tensor, key_pe: Tensor
+    ) -> Tuple[Tensor, Tensor]:
+        # Self attention block
+        if self.skip_first_layer_pe:
+            queries = self.self_attn(q=queries, k=queries, v=queries)
+        else:
+            q = queries + query_pe
+            attn_out = self.self_attn(q=q, k=q, v=queries)
+            queries = queries + attn_out
+        queries = self.norm1(queries)
+
+        # Cross attention block, tokens attending to image embedding
+        q = queries + query_pe
+        k = keys + key_pe
+        attn_out = self.cross_attn_token_to_image(q=q, k=k, v=keys)
+        queries = queries + attn_out
+        queries = self.norm2(queries)
+
+        # MLP block
+        mlp_out = self.mlp(queries)
+        queries = queries + mlp_out
+        queries = self.norm3(queries)
+
+        # Cross attention block, image embedding attending to tokens
+        q = queries + query_pe
+        k = keys + key_pe
+        attn_out = self.cross_attn_image_to_token(q=k, k=q, v=queries)
+        keys = keys + attn_out
+        keys = self.norm4(keys)
+
+        return queries, keys
+
+
+class Attention(nn.Module):
+    """
+    An attention layer that allows for downscaling the size of the embedding
+    after projection to queries, keys, and values.
+    """
+
+    def __init__(
+        self,
+        embedding_dim: int,
+        num_heads: int,
+        downsample_rate: int = 1,
+    ) -> None:
+        super().__init__()
+        self.embedding_dim = embedding_dim
+        self.internal_dim = embedding_dim // downsample_rate
+        self.num_heads = num_heads
+        assert self.internal_dim % num_heads == 0, "num_heads must divide embedding_dim."
+
+        self.q_proj = nn.Linear(embedding_dim, self.internal_dim)
+        self.k_proj = nn.Linear(embedding_dim, self.internal_dim)
+        self.v_proj = nn.Linear(embedding_dim, self.internal_dim)
+        self.out_proj = nn.Linear(self.internal_dim, embedding_dim)
+
+    def _separate_heads(self, x: Tensor, num_heads: int) -> Tensor:
+        b, n, c = x.shape
+        x = x.reshape(b, n, num_heads, c // num_heads)
+        return x.transpose(1, 2)  # B x N_heads x N_tokens x C_per_head
+
+    def _recombine_heads(self, x: Tensor) -> Tensor:
+        b, n_heads, n_tokens, c_per_head = x.shape
+        x = x.transpose(1, 2)
+        return x.reshape(b, n_tokens, n_heads * c_per_head)  # B x N_tokens x C
+
+    def forward(self, q: Tensor, k: Tensor, v: Tensor) -> Tensor:
+        # Input projections
+        q = self.q_proj(q)
+        k = self.k_proj(k)
+        v = self.v_proj(v)
+
+        # Separate into heads
+        q = self._separate_heads(q, self.num_heads)
+        k = self._separate_heads(k, self.num_heads)
+        v = self._separate_heads(v, self.num_heads)
+
+        # Attention
+        _, _, _, c_per_head = q.shape
+        attn = q @ k.permute(0, 1, 3, 2)  # B x N_heads x N_tokens x N_tokens
+        attn = attn / math.sqrt(c_per_head)
+        attn = torch.softmax(attn, dim=-1)
+
+        # Get output
+        out = attn @ v
+        out = self._recombine_heads(out)
+        out = self.out_proj(out)
+
+        return out
+
+
+
+class LayerNorm3d(nn.Module):
+    def __init__(self, num_channels: int, eps: float = 1e-6) -> None:
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(num_channels))
+        self.bias = nn.Parameter(torch.zeros(num_channels))
+        self.eps = eps
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        u = x.mean(1, keepdim=True)
+        s = (x - u).pow(2).mean(1, keepdim=True)
+        x = (x - u) / torch.sqrt(s + self.eps)
+        x = self.weight[:, None, None, None] * x + self.bias[:, None, None, None]
+        return x
+
+
+class MaskDecoder3D(nn.Module):
+    def __init__(
+        self,
+        *,
+        transformer_dim: int,
+        # transformer: nn.Module ,
+        num_multimask_outputs: int = 3,
+        activation: Type[nn.Module] = nn.GELU,
+        iou_head_depth: int = 3,
+        iou_head_hidden_dim: int = 256,
+    ) -> None:
+        """
+        Predicts masks given an image and prompt embeddings, using a
+        transformer architecture.
+
+        Arguments:
+          transformer_dim (int): the channel dimension of the transformer
+          transformer (nn.Module): the transformer used to predict masks
+          num_multimask_outputs (int): the number of masks to predict
+            when disambiguating masks
+          activation (nn.Module): the type of activation to use when
+            upscaling masks
+          iou_head_depth (int): the depth of the MLP used to predict
+            mask quality
+          iou_head_hidden_dim (int): the hidden dimension of the MLP
+            used to predict mask quality
+        """
+        super().__init__()
+        self.transformer_dim = transformer_dim
+        # self.transformer = transformer
+        self.transformer = TwoWayTransformer3D(
+                depth=2,
+                embedding_dim=self.transformer_dim,
+                mlp_dim=2048,
+                num_heads=8,
+            )
+
+        self.num_multimask_outputs = num_multimask_outputs
+
+        self.iou_token = nn.Embedding(1, transformer_dim)
+        self.num_mask_tokens = num_multimask_outputs + 1
+        self.mask_tokens = nn.Embedding(self.num_mask_tokens, transformer_dim)
+
+        self.output_upscaling = nn.Sequential(
+            nn.ConvTranspose3d(transformer_dim, transformer_dim // 4, kernel_size=2, stride=2),
+            LayerNorm3d(transformer_dim // 4),
+            activation(),
+            nn.ConvTranspose3d(transformer_dim // 4, transformer_dim // 8, kernel_size=2, stride=2),
+            activation(),
+        )
+        self.output_hypernetworks_mlps = nn.ModuleList(
+            [
+                MLP(transformer_dim, transformer_dim, transformer_dim // 8, 3)
+                for i in range(self.num_mask_tokens)
+            ]
+        )
+
+        self.iou_prediction_head = MLP(
+            transformer_dim, iou_head_hidden_dim, self.num_mask_tokens, iou_head_depth
+        )
+
+    def forward(
+        self,
+        image_embeddings: torch.Tensor,
+        image_pe: torch.Tensor,
+        sparse_prompt_embeddings: torch.Tensor,
+        dense_prompt_embeddings: torch.Tensor,
+        multimask_output: bool,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Predict masks given image and prompt embeddings.
+
+        Arguments:
+          image_embeddings (torch.Tensor): the embeddings from the image encoder
+          image_pe (torch.Tensor): positional encoding with the shape of image_embeddings
+          sparse_prompt_embeddings (torch.Tensor): the embeddings of the points and boxes
+          dense_prompt_embeddings (torch.Tensor): the embeddings of the mask inputs
+          multimask_output (bool): Whether to return multiple masks or a single
+            mask.
+
+        Returns:
+          torch.Tensor: batched predicted masks
+          torch.Tensor: batched predictions of mask quality
+        """
+        masks, iou_pred = self.predict_masks(
+            image_embeddings=image_embeddings,
+            image_pe=image_pe,
+            sparse_prompt_embeddings=sparse_prompt_embeddings,
+            dense_prompt_embeddings=dense_prompt_embeddings,
+        )
+
+        # Select the correct mask or masks for output
+        if multimask_output:
+            mask_slice = slice(1, None)
+        else:
+            mask_slice = slice(0, 1)
+        masks = masks[:, mask_slice, :, :]
+        iou_pred = iou_pred[:, mask_slice]
+
+        # Prepare output
+        return masks, iou_pred
+
+    def predict_masks(
+        self,
+        image_embeddings: torch.Tensor,
+        image_pe: torch.Tensor,
+        sparse_prompt_embeddings: torch.Tensor,
+        dense_prompt_embeddings: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Predicts masks. See 'forward' for more details."""
+        # Concatenate output tokens
+        output_tokens = torch.cat([self.iou_token.weight, self.mask_tokens.weight], dim=0)
+        output_tokens = output_tokens.unsqueeze(0).expand(sparse_prompt_embeddings.size(0), -1, -1)
+        tokens = torch.cat((output_tokens, sparse_prompt_embeddings), dim=1)
+
+        # Expand per-image data in batch direction to be per-mask
+        if image_embeddings.shape[0] != tokens.shape[0]:
+            src = torch.repeat_interleave(image_embeddings, tokens.shape[0], dim=0)
+        else:
+            src = image_embeddings
+        src = src + dense_prompt_embeddings
+        if image_pe.shape[0] != tokens.shape[0]:
+            pos_src = torch.repeat_interleave(image_pe, tokens.shape[0], dim=0)
+        else:
+            pos_src = image_pe
+        b, c, x, y, z = src.shape
+
+        # Run the transformer
+        # import IPython; IPython.embed()
+        hs, src = self.transformer(src, pos_src, tokens)
+        iou_token_out = hs[:, 0, :]
+        mask_tokens_out = hs[:, 1 : (1 + self.num_mask_tokens), :]
+
+        # Upscale mask embeddings and predict masks using the mask tokens
+        src = src.transpose(1, 2).view(b, c, x, y, z)
+        upscaled_embedding = self.output_upscaling(src)
+        hyper_in_list: List[torch.Tensor] = []
+        for i in range(self.num_mask_tokens):
+            hyper_in_list.append(self.output_hypernetworks_mlps[i](mask_tokens_out[:, i, :]))
+        hyper_in = torch.stack(hyper_in_list, dim=1)
+        b, c, x, y, z = upscaled_embedding.shape
+        masks = (hyper_in @ upscaled_embedding.view(b, c, x * y * z)).view(b, -1, x, y, z)
+
+        # Generate mask quality predictions
+        iou_pred = self.iou_prediction_head(iou_token_out)
+
+        return masks, iou_pred
+
+
+# Lightly adapted from
+# https://github.com/facebookresearch/MaskFormer/blob/main/mask_former/modeling/transformer/transformer_predictor.py # noqa
+class MLP(nn.Module):
+    def __init__(
+        self,
+        input_dim: int,
+        hidden_dim: int,
+        output_dim: int,
+        num_layers: int,
+        sigmoid_output: bool = False,
+    ) -> None:
+        super().__init__()
+        self.num_layers = num_layers
+        h = [hidden_dim] * (num_layers - 1)
+        self.layers = nn.ModuleList(
+            nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim])
+        )
+        self.sigmoid_output = sigmoid_output
+
+    def forward(self, x):
+        for i, layer in enumerate(self.layers):
+            x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x)
+        if self.sigmoid_output:
+            x = F.sigmoid(x)
+        return x
+

segment_anything/modeling/prompt_encoder.py

@@ -0,0 +1,227 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import numpy as np
+import torch
+from torch import nn
+
+from typing import Any, Optional, Tuple, Type
+
+from .common import LayerNorm2d
+
+
+class PromptEncoder(nn.Module):
+    def __init__(
+        self,
+        embed_dim: int,
+        image_embedding_size: Tuple[int, int],
+        input_image_size: Tuple[int, int],
+        mask_in_chans: int,
+        activation: Type[nn.Module] = nn.GELU,
+    ) -> None:
+        """
+        Encodes prompts for input to SAM's mask decoder.
+
+        Arguments:
+          embed_dim (int): The prompts' embedding dimension
+          image_embedding_size (tuple(int, int)): The spatial size of the
+            image embedding, as (H, W).
+          input_image_size (int): The padded size of the image as input
+            to the image encoder, as (H, W).
+          mask_in_chans (int): The number of hidden channels used for
+            encoding input masks.
+          activation (nn.Module): The activation to use when encoding
+            input masks.
+        """
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.input_image_size = input_image_size
+        self.image_embedding_size = image_embedding_size
+        self.pe_layer = PositionEmbeddingRandom(embed_dim // 2)
+
+        self.num_point_embeddings: int = 4  # pos/neg point + 2 box corners
+        point_embeddings = [nn.Embedding(1, embed_dim) for i in range(self.num_point_embeddings)]
+        self.point_embeddings = nn.ModuleList(point_embeddings)
+        self.not_a_point_embed = nn.Embedding(1, embed_dim)
+
+        self.mask_input_size = (4 * image_embedding_size[0], 4 * image_embedding_size[1])
+        self.mask_downscaling = nn.Sequential(
+            nn.Conv2d(1, mask_in_chans // 4, kernel_size=2, stride=2),
+            LayerNorm2d(mask_in_chans // 4),
+            activation(),
+            nn.Conv2d(mask_in_chans // 4, mask_in_chans, kernel_size=2, stride=2),
+            LayerNorm2d(mask_in_chans),
+            activation(),
+            nn.Conv2d(mask_in_chans, embed_dim, kernel_size=1),
+        )
+        self.no_mask_embed = nn.Embedding(1, embed_dim)
+
+    def get_dense_pe(self) -> torch.Tensor:
+        """
+        Returns the positional encoding used to encode point prompts,
+        applied to a dense set of points the shape of the image encoding.
+
+        Returns:
+          torch.Tensor: Positional encoding with shape
+            1x(embed_dim)x(embedding_h)x(embedding_w)
+        """
+        return self.pe_layer(self.image_embedding_size).unsqueeze(0)
+
+    def _embed_points(
+        self,
+        points: torch.Tensor,
+        labels: torch.Tensor,
+        pad: bool,
+    ) -> torch.Tensor:
+        """Embeds point prompts."""
+        points = points + 0.5  # Shift to center of pixel
+
+        if pad:
+            padding_point = torch.zeros((points.shape[0], 1, 2), device=points.device)
+            padding_label = -torch.ones((labels.shape[0], 1), device=labels.device)
+            points = torch.cat([points, padding_point], dim=1)  #B,N+1,2
+            labels = torch.cat([labels, padding_label], dim=1)
+
+
+        point_embedding = self.pe_layer.forward_with_coords(points, self.input_image_size)  #B,N+1,256
+        point_embedding[labels == -1] = 0.0
+
+        self.not_a_point_embed.weight = torch.nn.Parameter(self.not_a_point_embed.weight.to(point_embedding.dtype), requires_grad=True)  # todo
+        self.point_embeddings[0].weight = torch.nn.Parameter(self.point_embeddings[0].weight.to(point_embedding.dtype), requires_grad=True) #todo
+        self.point_embeddings[1].weight = torch.nn.Parameter(self.point_embeddings[1].weight.to(point_embedding.dtype), requires_grad=True) #todo
+
+        point_embedding[labels == -1] += self.not_a_point_embed.weight 
+        point_embedding[labels == 0] += self.point_embeddings[0].weight
+        point_embedding[labels == 1] += self.point_embeddings[1].weight
+        return point_embedding
+
+    def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
+        """Embeds box prompts."""
+
+        boxes = boxes + 0.5  # Shift to center of pixel
+        coords = boxes.reshape(-1, 2, 2)
+        corner_embedding = self.pe_layer.forward_with_coords(coords, self.input_image_size)
+        corner_embedding[:, 0, :] += self.point_embeddings[2].weight
+        corner_embedding[:, 1, :] += self.point_embeddings[3].weight
+        return corner_embedding
+
+    def _embed_masks(self, masks: torch.Tensor) -> torch.Tensor:
+        """Embeds mask inputs."""
+        mask_embedding = self.mask_downscaling(masks)
+        return mask_embedding
+
+    def _get_batch_size(
+        self,
+        points: Optional[Tuple[torch.Tensor, torch.Tensor]],
+        boxes: Optional[torch.Tensor],
+        masks: Optional[torch.Tensor],
+    ) -> int:
+        """
+        Gets the batch size of the output given the batch size of the input prompts.
+        """
+        if points is not None:
+            return points[0].shape[0]
+        elif boxes is not None:
+            return boxes.shape[0]
+        elif masks is not None:
+            return masks.shape[0]
+        else:
+            return 1
+
+    def _get_device(self) -> torch.device:
+        return self.point_embeddings[0].weight.device
+
+    def forward(
+        self,
+        points: Optional[Tuple[torch.Tensor, torch.Tensor]],
+        boxes: Optional[torch.Tensor],
+        masks: Optional[torch.Tensor],
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Embeds different types of prompts, returning both sparse and dense
+        embeddings.
+
+        Arguments:
+          points (tuple(torch.Tensor, torch.Tensor) or none): point coordinates
+            and labels to embed.
+          boxes (torch.Tensor or none): boxes to embed
+          masks (torch.Tensor or none): masks to embed
+
+        Returns:
+          torch.Tensor: sparse embeddings for the points and boxes, with shape
+            BxNx(embed_dim), where N is determined by the number of input points
+            and boxes.
+          torch.Tensor: dense embeddings for the masks, in the shape
+            Bx(embed_dim)x(embed_H)x(embed_W)
+        """
+        bs = self._get_batch_size(points, boxes, masks)
+        sparse_embeddings = torch.empty((bs, 0, self.embed_dim), device=self._get_device()) #B,0,256  空[]
+
+        if points is not None:
+            coords, labels = points     #coords:B,N,2  labels:B,N
+            point_embeddings = self._embed_points(coords, labels, pad=(boxes is None))
+            sparse_embeddings = torch.cat([sparse_embeddings, point_embeddings], dim=1)
+
+        if boxes is not None:
+            box_embeddings = self._embed_boxes(boxes)
+            sparse_embeddings = torch.cat([sparse_embeddings, box_embeddings], dim=1)
+
+        if masks is not None:
+            dense_embeddings = self._embed_masks(masks)
+        else:
+            dense_embeddings = self.no_mask_embed.weight.reshape(1, -1, 1, 1).expand(
+                bs, -1, self.image_embedding_size[0], self.image_embedding_size[1]
+            )
+
+        return sparse_embeddings, dense_embeddings
+
+
+class PositionEmbeddingRandom(nn.Module):
+    """
+    Positional encoding using random spatial frequencies.
+    """
+
+    def __init__(self, num_pos_feats: int = 64, scale: Optional[float] = None) -> None:
+        super().__init__()
+        if scale is None or scale <= 0.0:
+            scale = 1.0
+        self.register_buffer(
+            "positional_encoding_gaussian_matrix",
+            scale * torch.randn((2, num_pos_feats)),
+        )
+
+    def _pe_encoding(self, coords: torch.Tensor) -> torch.Tensor:
+        """Positionally encode points that are normalized to [0,1]."""
+        # assuming coords are in [0, 1]^2 square and have d_1 x ... x d_n x 2 shape
+        coords = 2 * coords - 1
+        coords = coords @ self.positional_encoding_gaussian_matrix.to(torch.float32)
+        coords = 2 * np.pi * coords
+        # outputs d_1 x ... x d_n x C shape
+        return torch.cat([torch.sin(coords), torch.cos(coords)], dim=-1)
+
+    def forward(self, size: Tuple[int, int]) -> torch.Tensor:
+        """Generate positional encoding for a grid of the specified size."""
+        h, w = size
+
+        device: Any = self.positional_encoding_gaussian_matrix.device
+        grid = torch.ones((h, w), device=device, dtype=torch.float32)
+        y_embed = grid.cumsum(dim=0) - 0.5
+        x_embed = grid.cumsum(dim=1) - 0.5
+        y_embed = y_embed / h
+        x_embed = x_embed / w
+
+        pe = self._pe_encoding(torch.stack([x_embed, y_embed], dim=-1))
+        return pe.permute(2, 0, 1)  # C x H x W
+
+    def forward_with_coords(
+        self, coords_input: torch.Tensor, image_size: Tuple[int, int]
+    ) -> torch.Tensor:
+        """Positionally encode points that are not normalized to [0,1]."""
+        coords = coords_input.clone()
+        coords[:, :, 0] = coords[:, :, 0] / image_size[1]
+        coords[:, :, 1] = coords[:, :, 1] / image_size[0]
+
+        return self._pe_encoding(coords.to(torch.float))  # B x N x C

segment_anything/modeling/prompt_encoder3D.py

@@ -0,0 +1,230 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import numpy as np
+import torch
+from torch import nn
+
+from typing import Any, Optional, Tuple, Type
+
+
+class LayerNorm3d(nn.Module):
+    def __init__(self, num_channels: int, eps: float = 1e-6) -> None:
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(num_channels))
+        self.bias = nn.Parameter(torch.zeros(num_channels))
+        self.eps = eps
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        u = x.mean(1, keepdim=True)
+        s = (x - u).pow(2).mean(1, keepdim=True)
+        x = (x - u) / torch.sqrt(s + self.eps)
+        x = self.weight[:, None, None, None] * x + self.bias[:, None, None, None]
+        return x
+
+
+class PromptEncoder3D(nn.Module):
+    def __init__(
+        self,
+        embed_dim: int,
+        image_embedding_size: Tuple[int, int, int],
+        input_image_size: Tuple[int, int, int],
+        mask_in_chans: int,
+        activation: Type[nn.Module] = nn.GELU,
+    ) -> None:
+        """
+        Encodes prompts for input to SAM's mask decoder.
+
+        Arguments:
+          embed_dim (int): The prompts' embedding dimension
+          image_embedding_size (tuple(int, int)): The spatial size of the
+            image embedding, as (H, W).
+          input_image_size (int): The padded size of the image as input
+            to the image encoder, as (H, W).
+          mask_in_chans (int): The number of hidden channels used for
+            encoding input masks.
+          activation (nn.Module): The activation to use when encoding
+            input masks.
+        """
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.input_image_size = input_image_size
+        self.image_embedding_size = image_embedding_size
+        self.pe_layer = PositionEmbeddingRandom3D(embed_dim // 3)
+
+        self.num_point_embeddings: int = 2  # pos/neg point
+        point_embeddings = [nn.Embedding(1, embed_dim) for i in range(self.num_point_embeddings)]
+        self.point_embeddings = nn.ModuleList(point_embeddings)
+        self.not_a_point_embed = nn.Embedding(1, embed_dim)
+
+        self.mask_input_size = (image_embedding_size[0], image_embedding_size[1], image_embedding_size[2])
+        self.mask_downscaling = nn.Sequential(
+            nn.Conv3d(1, mask_in_chans // 4, kernel_size=2, stride=2),
+            LayerNorm3d(mask_in_chans // 4),
+            activation(),
+            nn.Conv3d(mask_in_chans // 4, mask_in_chans, kernel_size=2, stride=2),
+            LayerNorm3d(mask_in_chans),
+            activation(),
+            nn.Conv3d(mask_in_chans, embed_dim, kernel_size=1),
+        )
+        self.no_mask_embed = nn.Embedding(1, embed_dim)
+
+    def get_dense_pe(self) -> torch.Tensor:
+        """
+        Returns the positional encoding used to encode point prompts,
+        applied to a dense set of points the shape of the image encoding.
+
+        Returns:
+          torch.Tensor: Positional encoding with shape
+            1x(embed_dim)x(embedding_h)x(embedding_w)
+        """
+        return self.pe_layer(self.image_embedding_size).unsqueeze(0)  # 1xXxYxZ
+
+    def _embed_points(
+        self,
+        points: torch.Tensor,
+        labels: torch.Tensor,
+        pad: bool,
+    ) -> torch.Tensor:
+        """Embeds point prompts."""
+        points = points + 0.5  # Shift to center of pixel
+        if pad:
+            padding_point = torch.zeros((points.shape[0], 1, 3), device=points.device)
+            padding_label = -torch.ones((labels.shape[0], 1), device=labels.device)
+            points = torch.cat([points, padding_point], dim=1)
+            labels = torch.cat([labels, padding_label], dim=1)
+        point_embedding = self.pe_layer.forward_with_coords(points, self.input_image_size)
+        point_embedding[labels == -1] = 0.0
+        point_embedding[labels == -1] += self.not_a_point_embed.weight
+        point_embedding[labels == 0] += self.point_embeddings[0].weight
+        point_embedding[labels == 1] += self.point_embeddings[1].weight
+        return point_embedding
+
+    def _embed_boxes(self, boxes: torch.Tensor) -> torch.Tensor:
+        """Embeds box prompts."""
+        boxes = boxes + 0.5  # Shift to center of pixel
+        coords = boxes.reshape(-1, 2, 2)
+        corner_embedding = self.pe_layer.forward_with_coords(coords, self.input_image_size)
+        corner_embedding[:, 0, :] += self.point_embeddings[2].weight
+        corner_embedding[:, 1, :] += self.point_embeddings[3].weight
+        return corner_embedding
+
+    def _embed_masks(self, masks: torch.Tensor) -> torch.Tensor:
+        """Embeds mask inputs."""
+        mask_embedding = self.mask_downscaling(masks)
+        return mask_embedding
+
+    def _get_batch_size(
+        self,
+        points: Optional[Tuple[torch.Tensor, torch.Tensor]],
+        boxes: Optional[torch.Tensor],
+        masks: Optional[torch.Tensor],
+    ) -> int:
+        """
+        Gets the batch size of the output given the batch size of the input prompts.
+        """
+        if points is not None:
+            return points[0].shape[0]
+        elif boxes is not None:
+            return boxes.shape[0]
+        elif masks is not None:
+            return masks.shape[0]
+        else:
+            return 1
+
+    def _get_device(self) -> torch.device:
+        return self.point_embeddings[0].weight.device
+
+    def forward(
+        self,
+        points: Optional[Tuple[torch.Tensor, torch.Tensor]],
+        boxes: Optional[torch.Tensor],
+        masks: Optional[torch.Tensor],
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Embeds different types of prompts, returning both sparse and dense
+        embeddings.
+
+        Arguments:
+          points (tuple(torch.Tensor, torch.Tensor) or none): point coordinates
+            and labels to embed.
+          boxes (torch.Tensor or none): boxes to embed
+          masks (torch.Tensor or none): masks to embed
+
+        Returns:
+          torch.Tensor: sparse embeddings for the points and boxes, with shape
+            BxNx(embed_dim), where N is determined by the number of input points
+            and boxes.
+          torch.Tensor: dense embeddings for the masks, in the shape
+            Bx(embed_dim)x(embed_H)x(embed_W)
+        """
+        bs = self._get_batch_size(points, boxes, masks)
+        sparse_embeddings = torch.empty((bs, 0, self.embed_dim), device=self._get_device())
+        if points is not None:
+            coords, labels = points
+            point_embeddings = self._embed_points(coords, labels, pad=(boxes is None))
+            sparse_embeddings = torch.cat([sparse_embeddings, point_embeddings], dim=1)
+        if boxes is not None:
+            box_embeddings = self._embed_boxes(boxes)
+            sparse_embeddings = torch.cat([sparse_embeddings, box_embeddings], dim=1)
+
+        if masks is not None:
+            dense_embeddings = self._embed_masks(masks)
+        else:
+            dense_embeddings = self.no_mask_embed.weight.reshape(1, -1, 1, 1, 1).expand(
+                bs, -1, self.image_embedding_size[0], self.image_embedding_size[1], self.image_embedding_size[2]
+            )
+
+        return sparse_embeddings, dense_embeddings
+
+
+class PositionEmbeddingRandom3D(nn.Module):
+    """
+    Positional encoding using random spatial frequencies.
+    """
+
+    def __init__(self, num_pos_feats: int = 64, scale: Optional[float] = None) -> None:
+        super().__init__()
+        if scale is None or scale <= 0.0:
+            scale = 1.0
+        self.register_buffer(
+            "positional_encoding_gaussian_matrix",
+            scale * torch.randn((3, num_pos_feats)),
+        )
+
+    def _pe_encoding(self, coords: torch.Tensor) -> torch.Tensor:
+        """Positionally encode points that are normalized to [0,1]."""
+        # assuming coords are in [0, 1]^2 square and have d_1 x ... x d_n x 2 shape
+        coords = 2 * coords - 1
+        coords = coords @ self.positional_encoding_gaussian_matrix
+        coords = 2 * np.pi * coords
+        # outputs d_1 x ... x d_n x C shape
+        return torch.cat([torch.sin(coords), torch.cos(coords), torch.sin(coords)], dim=-1)
+
+    def forward(self, size: Tuple[int, int, int]) -> torch.Tensor:
+        """Generate positional encoding for a grid of the specified size."""
+        x, y, z = size
+        device: Any = self.positional_encoding_gaussian_matrix.device
+        grid = torch.ones((x, y, z), device=device, dtype=torch.float32)
+        y_embed = grid.cumsum(dim=0) - 0.5
+        x_embed = grid.cumsum(dim=1) - 0.5
+        z_embed = grid.cumsum(dim=2) - 0.5
+        y_embed = y_embed / y
+        x_embed = x_embed / x
+        z_embed = z_embed / z
+
+        pe = self._pe_encoding(torch.stack([x_embed, y_embed, z_embed], dim=-1))
+        return pe.permute(3, 0, 1, 2)  # C x X x Y x Z
+
+    def forward_with_coords(
+        self, coords_input: torch.Tensor, image_size: Tuple[int, int, int]
+    ) -> torch.Tensor:
+        """Positionally encode points that are not normalized to [0,1]."""
+        coords = coords_input.clone()
+        coords[:, :, 0] = coords[:, :, 0] / image_size[0]
+        coords[:, :, 1] = coords[:, :, 1] / image_size[1]
+        coords[:, :, 2] = coords[:, :, 2] / image_size[2]
+        return self._pe_encoding(coords.to(torch.float))  # B x N x C

segment_anything/modeling/sam.py

@@ -0,0 +1,174 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from typing import Any, Dict, List, Tuple
+
+from .image_encoder import ImageEncoderViT
+from .mask_decoder import MaskDecoder
+from .prompt_encoder import PromptEncoder
+
+
+class Sam(nn.Module):
+    mask_threshold: float = 0.0
+    image_format: str = "RGB"
+
+    def __init__(
+        self,
+        image_encoder: ImageEncoderViT,
+        prompt_encoder: PromptEncoder,
+        mask_decoder: MaskDecoder,
+        pixel_mean: List[float] = [123.675, 116.28, 103.53],
+        pixel_std: List[float] = [58.395, 57.12, 57.375],
+    ) -> None:
+        """
+        SAM predicts object masks from an image and input prompts.
+
+        Arguments:
+          image_encoder (ImageEncoderViT): The backbone used to encode the
+            image into image embeddings that allow for efficient mask prediction.
+          prompt_encoder (PromptEncoder): Encodes various types of input prompts.
+          mask_decoder (MaskDecoder): Predicts masks from the image embeddings
+            and encoded prompts.
+          pixel_mean (list(float)): Mean values for normalizing pixels in the input image.
+          pixel_std (list(float)): Std values for normalizing pixels in the input image.
+        """
+        super().__init__()
+        self.image_encoder = image_encoder
+        self.prompt_encoder = prompt_encoder
+        self.mask_decoder = mask_decoder
+        self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False)
+        self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False)
+
+    @property
+    def device(self) -> Any:
+        return self.pixel_mean.device
+
+    @torch.no_grad()
+    def forward(
+        self,
+        batched_input: List[Dict[str, Any]],
+        multimask_output: bool,
+    ) -> List[Dict[str, torch.Tensor]]:
+        """
+        Predicts masks end-to-end from provided images and prompts.
+        If prompts are not known in advance, using SamPredictor is
+        recommended over calling the model directly.
+
+        Arguments:
+          batched_input (list(dict)): A list over input images, each a
+            dictionary with the following keys. A prompt key can be
+            excluded if it is not present.
+              'image': The image as a torch tensor in 3xHxW format,
+                already transformed for input to the model.
+              'original_size': (tuple(int, int)) The original size of
+                the image before transformation, as (H, W).
+              'point_coords': (torch.Tensor) Batched point prompts for
+                this image, with shape BxNx2. Already transformed to the
+                input frame of the model.
+              'point_labels': (torch.Tensor) Batched labels for point prompts,
+                with shape BxN.
+              'boxes': (torch.Tensor) Batched box inputs, with shape Bx4.
+                Already transformed to the input frame of the model.
+              'mask_inputs': (torch.Tensor) Batched mask inputs to the model,
+                in the form Bx1xHxW.
+          multimask_output (bool): Whether the model should predict multiple
+            disambiguating masks, or return a single mask.
+
+        Returns:
+          (list(dict)): A list over input images, where each element is
+            as dictionary with the following keys.
+              'masks': (torch.Tensor) Batched binary mask predictions,
+                with shape BxCxHxW, where B is the number of input prompts,
+                C is determined by multimask_output, and (H, W) is the
+                original size of the image.
+              'iou_predictions': (torch.Tensor) The model's predictions
+                of mask quality, in shape BxC.
+              'low_res_logits': (torch.Tensor) Low resolution logits with
+                shape BxCxHxW, where H=W=256. Can be passed as mask input
+                to subsequent iterations of prediction.
+        """
+
+        input_images = torch.stack([self.preprocess(x["image"]) for x in batched_input], dim=0)
+        image_embeddings = self.image_encoder(input_images)
+
+        outputs = []
+        for image_record, curr_embedding in zip(batched_input, image_embeddings):
+            if "point_coords" in image_record:
+                points = (image_record["point_coords"], image_record["point_labels"])
+            else:
+                points = None
+            sparse_embeddings, dense_embeddings = self.prompt_encoder(
+                points=points,
+                boxes=image_record.get("boxes", None),
+                masks=image_record.get("mask_inputs", None),
+            )
+            low_res_masks, iou_predictions = self.mask_decoder(
+                image_embeddings=curr_embedding.unsqueeze(0),
+                image_pe=self.prompt_encoder.get_dense_pe(),
+                sparse_prompt_embeddings=sparse_embeddings,
+                dense_prompt_embeddings=dense_embeddings,
+                multimask_output=multimask_output,
+            )
+            masks = self.postprocess_masks(
+                low_res_masks,
+                input_size=image_record["image"].shape[-2:],
+                original_size=image_record["original_size"],
+            )
+            masks = masks > self.mask_threshold
+            outputs.append(
+                {
+                    "masks": masks,
+                    "iou_predictions": iou_predictions,
+                    "low_res_logits": low_res_masks,
+                }
+            )
+        return outputs
+
+    def postprocess_masks(
+        self,
+        masks: torch.Tensor,
+        input_size: Tuple[int, ...],
+        original_size: Tuple[int, ...],
+    ) -> torch.Tensor:
+        """
+        Remove padding and upscale masks to the original image size.
+
+        Arguments:
+          masks (torch.Tensor): Batched masks from the mask_decoder,
+            in BxCxHxW format.
+          input_size (tuple(int, int)): The size of the image input to the
+            model, in (H, W) format. Used to remove padding.
+          original_size (tuple(int, int)): The original size of the image
+            before resizing for input to the model, in (H, W) format.
+
+        Returns:
+          (torch.Tensor): Batched masks in BxCxHxW format, where (H, W)
+            is given by original_size.
+        """
+        masks = F.interpolate(
+            masks,
+            (self.image_encoder.img_size, self.image_encoder.img_size),
+            mode="bilinear",
+            align_corners=False,
+        )
+        masks = masks[..., : input_size[0], : input_size[1]]
+        masks = F.interpolate(masks, original_size, mode="bilinear", align_corners=False)
+        return masks
+
+    def preprocess(self, x: torch.Tensor) -> torch.Tensor:
+        """Normalize pixel values and pad to a square input."""
+        # Normalize colors
+        x = (x - self.pixel_mean) / self.pixel_std
+        # Pad
+        h, w = x.shape[-2:]
+        padh = self.image_encoder.img_size - h
+        padw = self.image_encoder.img_size - w
+        x = F.pad(x, (0, padw, 0, padh))
+        return x

segment_anything/modeling/sam3D.py

@@ -0,0 +1,176 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from typing import Any, Dict, List, Tuple
+
+from .image_encoder3D import ImageEncoderViT3D
+from .mask_decoder3D import MaskDecoder3D
+from .prompt_encoder3D import PromptEncoder3D
+
+
+class Sam3D(nn.Module):
+    mask_threshold: float = 0.0
+    image_format: str = "L"
+
+    def __init__(
+        self,
+        image_encoder: ImageEncoderViT3D,
+        prompt_encoder: PromptEncoder3D,
+        mask_decoder: MaskDecoder3D,
+        pixel_mean: List[float] = [123.675],
+        pixel_std: List[float] = [58.395],
+    ) -> None:
+        """
+        SAM predicts object masks from an image and input prompts.
+
+        Arguments:
+          image_encoder (ImageEncoderViT): The backbone used to encode the
+            image into image embeddings that allow for efficient mask prediction.
+          prompt_encoder (PromptEncoder): Encodes various types of input prompts.
+          mask_decoder (MaskDecoder): Predicts masks from the image embeddings
+            and encoded prompts.
+          pixel_mean (list(float)): Mean values for normalizing pixels in the input image.
+          pixel_std (list(float)): Std values for normalizing pixels in the input image.
+        """
+        super().__init__()
+        self.image_encoder = image_encoder
+        self.prompt_encoder = prompt_encoder
+        self.mask_decoder = mask_decoder
+        self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False)
+        self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False)
+
+    @property
+    def device(self) -> Any:
+        return self.pixel_mean.device
+
+    @torch.no_grad()
+    def forward(
+        self,
+        batched_input: List[Dict[str, Any]],
+        multimask_output: bool,
+    ) -> List[Dict[str, torch.Tensor]]:
+        """
+        Predicts masks end-to-end from provided images and prompts.
+        If prompts are not known in advance, using SamPredictor is
+        recommended over calling the model directly.
+
+        Arguments:
+          batched_input (list(dict)): A list over input images, each a
+            dictionary with the following keys. A prompt key can be
+            excluded if it is not present.
+              'image': The image as a torch tensor in 3xHxW format,
+                already transformed for input to the model.
+              'original_size': (tuple(int, int)) The original size of
+                the image before transformation, as (H, W).
+              'point_coords': (torch.Tensor) Batched point prompts for
+                this image, with shape BxNx2. Already transformed to the
+                input frame of the model.
+              'point_labels': (torch.Tensor) Batched labels for point prompts,
+                with shape BxN.
+              'boxes': (torch.Tensor) Batched box inputs, with shape Bx4.
+                Already transformed to the input frame of the model.
+              'mask_inputs': (torch.Tensor) Batched mask inputs to the model,
+                in the form Bx1xHxW.
+          multimask_output (bool): Whether the model should predict multiple
+            disambiguating masks, or return a single mask.
+
+        Returns:
+          (list(dict)): A list over input images, where each element is
+            as dictionary with the following keys.
+              'masks': (torch.Tensor) Batched binary mask predictions,
+                with shape BxCxHxW, where B is the number of input prompts,
+                C is determined by multimask_output, and (H, W) is the
+                original size of the image.
+              'iou_predictions': (torch.Tensor) The model's predictions
+                of mask quality, in shape BxC.
+              'low_res_logits': (torch.Tensor) Low resolution logits with
+                shape BxCxHxW, where H=W=256. Can be passed as mask input
+                to subsequent iterations of prediction.
+        """
+        input_images = torch.stack([self.preprocess(x["image"]) for x in batched_input], dim=0)
+        image_embeddings = self.image_encoder(input_images)
+
+        outputs = []
+        for image_record, curr_embedding in zip(batched_input, image_embeddings):
+            if "point_coords" in image_record:
+                points = (image_record["point_coords"], image_record["point_labels"])
+            else:
+                points = None
+            sparse_embeddings, dense_embeddings = self.prompt_encoder(
+                points=points,
+                boxes=image_record.get("boxes", None),
+                masks=image_record.get("mask_inputs", None),
+            )
+            low_res_masks, iou_predictions = self.mask_decoder(
+                image_embeddings=curr_embedding.unsqueeze(0),
+                image_pe=self.prompt_encoder.get_dense_pe(),
+                sparse_prompt_embeddings=sparse_embeddings,
+                dense_prompt_embeddings=dense_embeddings,
+                multimask_output=multimask_output,
+            )
+            masks = self.postprocess_masks(
+                low_res_masks,
+                input_size=image_record["image"].shape[-3:],
+                original_size=image_record["original_size"],
+            )
+            masks = masks > self.mask_threshold
+            outputs.append(
+                {
+                    "masks": masks,
+                    "iou_predictions": iou_predictions,
+                    "low_res_logits": low_res_masks,
+                }
+            )
+        return outputs
+
+    def postprocess_masks(
+        self,
+        masks: torch.Tensor,
+        input_size: Tuple[int, ...],
+        original_size: Tuple[int, ...],
+    ) -> torch.Tensor:
+        """
+        Remove padding and upscale masks to the original image size.
+
+        Arguments:
+          masks (torch.Tensor): Batched masks from the mask_decoder,
+            in BxCxHxW format.
+          input_size (tuple(int, int)): The size of the image input to the
+            model, in (H, W) format. Used to remove padding.
+          original_size (tuple(int, int)): The original size of the image
+            before resizing for input to the model, in (H, W) format.
+
+        Returns:
+          (torch.Tensor): Batched masks in BxCxHxW format, where (H, W)
+            is given by original_size.
+        """
+        masks = F.interpolate(
+            masks,
+            (self.image_encoder.img_size, self.image_encoder.img_size, self.image_encoder.img_size),
+            mode="bilinear",
+            align_corners=False,
+        )
+        masks = masks[..., : input_size[0], : input_size[1], : input_size[2]]
+        masks = F.interpolate(masks, original_size, mode="bilinear", align_corners=False)
+        return masks
+
+    def preprocess(self, x: torch.Tensor) -> torch.Tensor:
+        """Normalize pixel values and pad to a square input."""
+        # Normalize colors
+        x = (x - self.pixel_mean) / self.pixel_std
+
+        # Pad
+        d, h, w = x.shape[-3:]
+        padd = self.image_encoder.img_size - d
+        padh = self.image_encoder.img_size - h
+        padw = self.image_encoder.img_size - w
+        x = F.pad(x, (0, padw, 0, padh, 0, padd))
+        return x
+

segment_anything/modeling/sam_model.py

@@ -0,0 +1,106 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+import torch
+from torch import nn
+from torch.nn import functional as F
+from typing import Any, Dict, List, Tuple
+from .image_encoder import ImageEncoderViT
+from .mask_decoder import MaskDecoder
+from .prompt_encoder import PromptEncoder
+
+
+class Sam(nn.Module):
+    mask_threshold: float = 0.0
+    image_format: str = "RGB"
+
+    def __init__(
+        self,
+        image_encoder: ImageEncoderViT,
+        prompt_encoder: PromptEncoder,
+        mask_decoder: MaskDecoder,
+        pixel_mean: List[float] = [123.675, 116.28, 103.53],
+        pixel_std: List[float] = [58.395, 57.12, 57.375],
+    ) -> None:
+        """
+        SAM predicts object masks from an image and input prompts.
+
+        Arguments:
+          image_encoder (ImageEncoderViT): The backbone used to encode the
+            image into image embeddings that allow for efficient mask prediction.
+          prompt_encoder (PromptEncoder): Encodes various types of input prompts.
+          mask_decoder (MaskDecoder): Predicts masks from the image embeddings
+            and encoded prompts.
+          pixel_mean (list(float)): Mean values for normalizing pixels in the input image.
+          pixel_std (list(float)): Std values for normalizing pixels in the input image.
+        """
+        super().__init__()
+        self.image_encoder = image_encoder
+        self.prompt_encoder = prompt_encoder
+        self.mask_decoder = mask_decoder
+        self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False)
+        self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False)
+
+    @property
+    def device(self) -> Any:
+        return self.pixel_mean.device
+
+    def forward(self, batched_input: Dict[str, Any], multimask_output: bool) -> List[Dict[str, torch.Tensor]]:
+
+        input_images = batched_input.get("image")
+        image_embeddings = self.image_encoder(input_images)
+
+        if "point_coords" in batched_input and batched_input["point_coords"] != None:
+            points = (batched_input["point_coords"], batched_input["point_labels"])
+        else:
+            points = None
+
+        sparse_embeddings, dense_embeddings = self.prompt_encoder(
+            points=points,
+            boxes=batched_input.get("boxes", None),
+            masks=batched_input.get("mask_inputs", None),
+        )  # sparse_embeddings:[2, 3, 256],  dense_embeddings:[2, 256, 64, 64]
+
+        low_res_masks, iou_predictions = self.mask_decoder(
+            image_embeddings=image_embeddings,
+            image_pe=self.prompt_encoder.get_dense_pe(),  # 1x(256)x(64)x(64)
+            sparse_prompt_embeddings=sparse_embeddings,
+            dense_prompt_embeddings=dense_embeddings,
+            multimask_output=multimask_output,
+        )
+
+        masks = self.postprocess_masks(
+            low_res_masks,
+            input_size=batched_input["image"].shape[-2:],
+            original_size=batched_input["original_size"],
+        )
+
+        outputs = {
+                    "masks": masks,
+                    "iou_predictions": iou_predictions,
+                    "low_res_logits": low_res_masks,
+                }
+
+        return outputs
+
+    def postprocess_masks(self,masks: torch.Tensor, input_size: Tuple[int, ...],original_size: Tuple[int, ...],) -> torch.Tensor:
+        masks = F.interpolate(
+            masks,
+            (self.image_encoder.img_size, self.image_encoder.img_size), mode="bilinear", align_corners=False,)  #[1,1024,1024]
+
+        masks = masks[..., : input_size[0], : input_size[1]]
+        masks = F.interpolate(masks, original_size, mode="bilinear", align_corners=False)
+        return masks
+
+    def preprocess(self, x: torch.Tensor) -> torch.Tensor:
+        """Normalize pixel values and pad to a square input."""
+        # Normalize colors
+        x = (x - self.pixel_mean) / self.pixel_std
+        # Pad
+        h, w = x.shape[-2:]
+        padh = self.image_encoder.img_size - h
+        padw = self.image_encoder.img_size - w
+        x = F.pad(x, (0, padw, 0, padh))
+        return x

segment_anything/modeling/transformer.py

@@ -0,0 +1,244 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torch import Tensor, nn
+
+import math
+from typing import Tuple, Type
+
+from .common import MLPBlock
+
+
+class TwoWayTransformer(nn.Module):
+    def __init__(
+        self,
+        depth: int,
+        embedding_dim: int,
+        num_heads: int,
+        mlp_dim: int,
+        activation: Type[nn.Module] = nn.ReLU,
+        attention_downsample_rate: int = 2,
+    ) -> None:
+        """
+        A transformer decoder that attends to an input image using
+        queries whose positional embedding is supplied.
+
+        Args:
+          depth (int): number of layers in the transformer
+          embedding_dim (int): the channel dimension for the input embeddings
+          num_heads (int): the number of heads for multihead attention. Must
+            divide embedding_dim
+          mlp_dim (int): the channel dimension internal to the MLP block
+          activation (nn.Module): the activation to use in the MLP block
+        """
+        super().__init__()
+        self.depth = depth
+        self.embedding_dim = embedding_dim
+        self.num_heads = num_heads
+        self.mlp_dim = mlp_dim
+        self.layers = nn.ModuleList()
+
+        for i in range(depth):
+            self.layers.append(
+                TwoWayAttentionBlock(
+                    embedding_dim=embedding_dim,
+                    num_heads=num_heads,
+                    mlp_dim=mlp_dim,
+                    activation=activation,
+                    attention_downsample_rate=attention_downsample_rate,
+                    skip_first_layer_pe=(i == 0),
+                )
+            )
+
+        self.final_attn_token_to_image = Attention(
+            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
+        )
+        self.norm_final_attn = nn.LayerNorm(embedding_dim)
+
+    def forward(
+        self,
+        image_embedding: Tensor,
+        image_pe: Tensor,
+        point_embedding: Tensor,
+    ) -> Tuple[Tensor, Tensor]:
+        """
+        Args:
+          image_embedding (torch.Tensor): image to attend to. Should be shape
+            B x embedding_dim x h x w for any h and w.
+          image_pe (torch.Tensor): the positional encoding to add to the image. Must
+            have the same shape as image_embedding.
+          point_embedding (torch.Tensor): the embedding to add to the query points.
+            Must have shape B x N_points x embedding_dim for any N_points.
+
+        Returns:
+          torch.Tensor: the processed point_embedding
+          torch.Tensor: the processed image_embedding
+        """
+        # BxCxHxW -> BxHWxC == B x N_image_tokens x C
+        bs, c, h, w = image_embedding.shape
+        image_embedding = image_embedding.flatten(2).permute(0, 2, 1)
+        image_pe = image_pe.flatten(2).permute(0, 2, 1)
+
+        # Prepare queries
+        queries = point_embedding
+        keys = image_embedding
+
+        # Apply transformer blocks and final layernorm
+        for layer in self.layers:
+            queries, keys = layer(
+                queries=queries,
+                keys=keys,
+                query_pe=point_embedding,
+                key_pe=image_pe,
+            )
+
+        # Apply the final attention layer from the points to the image
+        q = queries + point_embedding
+        k = keys + image_pe
+        attn_out = self.final_attn_token_to_image(q=q, k=k, v=keys)
+        queries = queries + attn_out
+        queries = self.norm_final_attn(queries)
+
+        return queries, keys
+
+
+class TwoWayAttentionBlock(nn.Module):
+    def     __init__(
+        self,
+        embedding_dim: int,
+        num_heads: int,
+        mlp_dim: int = 2048,
+        activation: Type[nn.Module] = nn.ReLU,
+        attention_downsample_rate: int = 2,
+        skip_first_layer_pe: bool = False,
+    ) -> None:
+        """
+        A transformer block with four layers: (1) self-attention of sparse
+        inputs, (2) cross attention of sparse inputs to dense inputs, (3) mlp
+        block on sparse inputs, and (4) cross attention of dense inputs to sparse
+        inputs.
+
+        Arguments:
+          embedding_dim (int): the channel dimension of the embeddings
+          num_heads (int): the number of heads in the attention layers
+          mlp_dim (int): the hidden dimension of the mlp block
+          activation (nn.Module): the activation of the mlp block
+          skip_first_layer_pe (bool): skip the PE on the first layer
+        """
+        super().__init__()
+        self.self_attn = Attention(embedding_dim, num_heads)
+        self.norm1 = nn.LayerNorm(embedding_dim)
+
+        self.cross_attn_token_to_image = Attention(
+            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
+        )
+        self.norm2 = nn.LayerNorm(embedding_dim)
+
+        self.mlp = MLPBlock(embedding_dim, mlp_dim, activation)
+        self.norm3 = nn.LayerNorm(embedding_dim)
+
+        self.norm4 = nn.LayerNorm(embedding_dim)
+        self.cross_attn_image_to_token = Attention(
+            embedding_dim, num_heads, downsample_rate=attention_downsample_rate
+        )
+
+        self.skip_first_layer_pe = skip_first_layer_pe
+
+    def forward(
+        self, queries: Tensor, keys: Tensor, query_pe: Tensor, key_pe: Tensor
+    ) -> Tuple[Tensor, Tensor]:
+        # Self attention block
+        if self.skip_first_layer_pe:
+            queries = self.self_attn(q=queries, k=queries, v=queries)
+        else:
+            q = queries + query_pe
+            attn_out = self.self_attn(q=q, k=q, v=queries)
+            queries = queries + attn_out
+        queries = self.norm1(queries)
+
+        # Cross attention block, tokens attending to image embedding
+        q = queries + query_pe
+        k = keys + key_pe
+        attn_out = self.cross_attn_token_to_image(q=q, k=k, v=keys)
+        queries = queries + attn_out
+        queries = self.norm2(queries)
+
+        # MLP block
+        mlp_out = self.mlp(queries)
+        queries = queries + mlp_out
+        queries = self.norm3(queries)
+
+        # Cross attention block, image embedding attending to tokens
+        q = queries + query_pe
+        k = keys + key_pe
+        attn_out = self.cross_attn_image_to_token(q=k, k=q, v=queries)
+        keys = keys + attn_out
+        keys = self.norm4(keys)
+
+        return queries, keys
+
+
+class Attention(nn.Module):
+    """
+    An attention layer that allows for downscaling the size of the embedding
+    after projection to queries, keys, and values.
+    """
+
+    def __init__(
+        self,
+        embedding_dim: int,
+        num_heads: int,
+        downsample_rate: int = 1,
+    ) -> None:
+        super().__init__()
+        self.embedding_dim = embedding_dim
+        self.internal_dim = embedding_dim // downsample_rate
+        self.num_heads = num_heads
+        assert self.internal_dim % num_heads == 0, "num_heads must divide embedding_dim."
+
+        self.q_proj = nn.Linear(embedding_dim, self.internal_dim)
+        self.k_proj = nn.Linear(embedding_dim, self.internal_dim)
+        self.v_proj = nn.Linear(embedding_dim, self.internal_dim)
+        self.out_proj = nn.Linear(self.internal_dim, embedding_dim)
+
+    def _separate_heads(self, x: Tensor, num_heads: int) -> Tensor:
+        b, n, c = x.shape
+        x = x.reshape(b, n, num_heads, c // num_heads)
+        return x.transpose(1, 2)  # B x N_heads x N_tokens x C_per_head
+
+    def _recombine_heads(self, x: Tensor) -> Tensor:
+        b, n_heads, n_tokens, c_per_head = x.shape
+        x = x.transpose(1, 2)
+        return x.reshape(b, n_tokens, n_heads * c_per_head)  # B x N_tokens x C
+
+    def forward(self, q: Tensor, k: Tensor, v: Tensor) -> Tensor:
+        # Input projections
+        q = self.q_proj(q.to(self.q_proj.weight.dtype)) #todo
+        k = self.k_proj(k.to(self.k_proj.weight.dtype)) #todo
+        v = self.v_proj(v.to(self.v_proj.weight.dtype)) #todo
+
+        # q = self.q_proj(q)
+        # k = self.k_proj(k)
+        # v = self.v_proj(v)
+
+        # Separate into heads
+        q = self._separate_heads(q, self.num_heads)
+        k = self._separate_heads(k, self.num_heads)
+        v = self._separate_heads(v, self.num_heads)
+
+        # Attention
+        _, _, _, c_per_head = q.shape
+        attn = q @ k.permute(0, 1, 3, 2)  # B x N_heads x N_tokens x N_tokens
+        attn = attn / math.sqrt(c_per_head)
+        attn = torch.softmax(attn, dim=-1)
+
+        # Get output
+        out = attn @ v
+        out = self._recombine_heads(out)
+        out = self.out_proj(out)
+
+        return out

segment_anything/predictor.py

@@ -0,0 +1,271 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import numpy as np
+import torch
+
+from segment_anything.modeling import Sam
+
+from typing import Optional, Tuple
+
+from .utils.transforms import ResizeLongestSide
+
+
+class SamPredictor:
+    def __init__(
+        self,
+        sam_model: Sam,
+    ) -> None:
+        """
+        Uses SAM to calculate the image embedding for an image, and then
+        allow repeated, efficient mask prediction given prompts.
+
+        Arguments:
+          sam_model (Sam): The model to use for mask prediction.
+        """
+        super().__init__()
+        self.model = sam_model
+        self.transform = ResizeLongestSide(sam_model.image_encoder.img_size)
+        self.reset_image()
+
+    def set_image(
+        self,
+        image: np.ndarray,
+        image_format: str = "RGB",
+    ) -> None:
+        """
+        Calculates the image embeddings for the provided image, allowing
+        masks to be predicted with the 'predict' method.
+
+        Arguments:
+          image (np.ndarray): The image for calculating masks. Expects an
+            image in HWC uint8 format, with pixel values in [0, 255].
+          image_format (str): The color format of the image, in ['RGB', 'BGR'].
+        """
+        assert image_format in [
+            "RGB",
+            "BGR",
+        ], f"image_format must be in ['RGB', 'BGR'], is {image_format}."
+        if image_format != self.model.image_format:
+            image = image[..., ::-1]
+
+        # Transform the image to the form expected by the model
+        input_image = self.transform.apply_image(image)
+        input_image_torch = torch.as_tensor(input_image, device=self.device)
+        input_image_torch = input_image_torch.permute(2, 0, 1).contiguous()[None, :, :, :]
+
+        self.set_torch_image(input_image_torch, image.shape[:2])
+
+    @torch.no_grad()
+    def set_torch_image(
+        self,
+        transformed_image: torch.Tensor,
+        original_image_size: Tuple[int, ...],
+    ) -> None:
+        """
+        Calculates the image embeddings for the provided image, allowing
+        masks to be predicted with the 'predict' method. Expects the input
+        image to be already transformed to the format expected by the model.
+
+        Arguments:
+          transformed_image (torch.Tensor): The input image, with shape
+            1x3xHxW, which has been transformed with ResizeLongestSide.
+          original_image_size (tuple(int, int)): The size of the image
+            before transformation, in (H, W) format.
+        """
+        assert (
+            len(transformed_image.shape) == 4
+            and transformed_image.shape[1] == 3
+            and max(*transformed_image.shape[2:]) == self.model.image_encoder.img_size
+        ), f"set_torch_image input must be BCHW with long side {self.model.image_encoder.img_size}."
+        self.reset_image()
+
+        self.original_size = original_image_size
+        self.input_size = tuple(transformed_image.shape[-2:])
+        input_image = self.model.preprocess(transformed_image)
+        self.features = self.model.image_encoder(input_image)
+        self.is_image_set = True
+
+    def predict(
+        self,
+        point_coords: Optional[np.ndarray] = None,
+        point_labels: Optional[np.ndarray] = None,
+        box: Optional[np.ndarray] = None,
+        mask_input: Optional[np.ndarray] = None,
+        multimask_output: bool = True,
+        return_logits: bool = False,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Predict masks for the given input prompts, using the currently set image.
+
+        Arguments:
+          point_coords (np.ndarray or None): A Nx2 array of point prompts to the
+            model. Each point is in (X,Y) in pixels.
+          point_labels (np.ndarray or None): A length N array of labels for the
+            point prompts. 1 indicates a foreground point and 0 indicates a
+            background point.
+          box (np.ndarray or None): A length 4 array given a box prompt to the
+            model, in XYXY format.
+          mask_input (np.ndarray): A low resolution mask input to the model, typically
+            coming from a previous prediction iteration. Has form 1xHxW, where
+            for SAM, H=W=256.
+          multimask_output (bool): If true, the model will return three masks.
+            For ambiguous input prompts (such as a single click), this will often
+            produce better masks than a single prediction. If only a single
+            mask is needed, the model's predicted quality score can be used
+            to select the best mask. For non-ambiguous prompts, such as multiple
+            input prompts, multimask_output=False can give better results.
+          return_logits (bool): If true, returns un-thresholded masks logits
+            instead of a binary mask.
+
+        Returns:
+          (np.ndarray): The output masks in CxHxW format, where C is the
+            number of masks, and (H, W) is the original image size.
+          (np.ndarray): An array of length C containing the model's
+            predictions for the quality of each mask.
+          (np.ndarray): An array of shape CxHxW, where C is the number
+            of masks and H=W=256. These low resolution logits can be passed to
+            a subsequent iteration as mask input.
+        """
+        if not self.is_image_set:
+            raise RuntimeError("An image must be set with .set_image(...) before mask prediction.")
+
+        # Transform input prompts
+        coords_torch, labels_torch, box_torch, mask_input_torch = None, None, None, None
+        if point_coords is not None:
+            assert (
+                point_labels is not None
+            ), "point_labels must be supplied if point_coords is supplied."
+
+            point_coords = self.transform.apply_coords(point_coords, self.original_size)
+            coords_torch = torch.as_tensor(point_coords, dtype=torch.float, device=self.device)
+            labels_torch = torch.as_tensor(point_labels, dtype=torch.int, device=self.device)
+            coords_torch, labels_torch = coords_torch[None, :, :], labels_torch[None, :]
+
+        if box is not None:
+            box = self.transform.apply_boxes(box, self.original_size)
+            box_torch = torch.as_tensor(box, dtype=torch.float, device=self.device)
+            box_torch = box_torch[None, :]
+        if mask_input is not None:
+            mask_input_torch = torch.as_tensor(mask_input, dtype=torch.float, device=self.device)
+            mask_input_torch = mask_input_torch[None, :, :, :]
+
+        masks, iou_predictions, low_res_masks = self.predict_torch(
+            coords_torch,
+            labels_torch,
+            box_torch,
+            mask_input_torch,
+            multimask_output,
+            return_logits=return_logits,
+        )
+
+        masks = masks[0].detach().cpu().numpy()
+        iou_predictions = iou_predictions[0].detach().cpu().numpy()
+        low_res_masks = low_res_masks[0].detach().cpu().numpy()
+        return masks, iou_predictions, low_res_masks
+
+    @torch.no_grad()
+    def predict_torch(
+        self,
+        point_coords: Optional[torch.Tensor],
+        point_labels: Optional[torch.Tensor],
+        boxes: Optional[torch.Tensor] = None,
+        mask_input: Optional[torch.Tensor] = None,
+        multimask_output: bool = True,
+        return_logits: bool = False,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Predict masks for the given input prompts, using the currently set image.
+        Input prompts are batched torch tensors and are expected to already be
+        transformed to the input frame using ResizeLongestSide.
+
+        Arguments:
+          point_coords (torch.Tensor or None): A BxNx2 array of point prompts to the
+            model. Each point is in (X,Y) in pixels.
+          point_labels (torch.Tensor or None): A BxN array of labels for the
+            point prompts. 1 indicates a foreground point and 0 indicates a
+            background point.
+          boxes (np.ndarray or None): A Bx4 array given a box prompt to the
+            model, in XYXY format.
+          mask_input (np.ndarray): A low resolution mask input to the model, typically
+            coming from a previous prediction iteration. Has form Bx1xHxW, where
+            for SAM, H=W=256. Masks returned by a previous iteration of the
+            predict method do not need further transformation.
+          multimask_output (bool): If true, the model will return three masks.
+            For ambiguous input prompts (such as a single click), this will often
+            produce better masks than a single prediction. If only a single
+            mask is needed, the model's predicted quality score can be used
+            to select the best mask. For non-ambiguous prompts, such as multiple
+            input prompts, multimask_output=False can give better results.
+          return_logits (bool): If true, returns un-thresholded masks logits
+            instead of a binary mask.
+
+        Returns:
+          (torch.Tensor): The output masks in BxCxHxW format, where C is the
+            number of masks, and (H, W) is the original image size.
+          (torch.Tensor): An array of shape BxC containing the model's
+            predictions for the quality of each mask.
+          (torch.Tensor): An array of shape BxCxHxW, where C is the number
+            of masks and H=W=256. These low res logits can be passed to
+            a subsequent iteration as mask input.
+        """
+        if not self.is_image_set:
+            raise RuntimeError("An image must be set with .set_image(...) before mask prediction.")
+
+        if point_coords is not None:
+            points = (point_coords, point_labels)
+        else:
+            points = None
+
+        # Embed prompts
+        sparse_embeddings, dense_embeddings = self.model.prompt_encoder(
+            points=points,
+            boxes=boxes,
+            masks=mask_input,
+        )
+
+        # Predict masks
+        low_res_masks, iou_predictions = self.model.mask_decoder(
+            image_embeddings=self.features,
+            image_pe=self.model.prompt_encoder.get_dense_pe(),
+            sparse_prompt_embeddings=sparse_embeddings,
+            dense_prompt_embeddings=dense_embeddings,
+            multimask_output=multimask_output,
+        )
+
+        # Upscale the masks to the original image resolution
+        masks = self.model.postprocess_masks(low_res_masks, self.input_size, self.original_size)
+
+        if not return_logits:
+            masks = masks > self.model.mask_threshold
+
+        return masks, iou_predictions, low_res_masks
+
+    def get_image_embedding(self) -> torch.Tensor:
+        """
+        Returns the image embeddings for the currently set image, with
+        shape 1xCxHxW, where C is the embedding dimension and (H,W) are
+        the embedding spatial dimension of SAM (typically C=256, H=W=64).
+        """
+        if not self.is_image_set:
+            raise RuntimeError(
+                "An image must be set with .set_image(...) to generate an embedding."
+            )
+        assert self.features is not None, "Features must exist if an image has been set."
+        return self.features
+
+    @property
+    def device(self) -> torch.device:
+        return self.model.device
+
+    def reset_image(self) -> None:
+        """Resets the currently set image."""
+        self.is_image_set = False
+        self.features = None
+        self.orig_h = None
+        self.orig_w = None
+        self.input_h = None
+        self.input_w = None

segment_anything/utils/__init__.py

@@ -0,0 +1 @@
+from .transforms3D import *

segment_anything/utils/amg.py

@@ -0,0 +1,346 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import numpy as np
+import torch
+
+import math
+from copy import deepcopy
+from itertools import product
+from typing import Any, Dict, Generator, ItemsView, List, Tuple
+
+
+class MaskData:
+    """
+    A structure for storing masks and their related data in batched format.
+    Implements basic filtering and concatenation.
+    """
+
+    def __init__(self, **kwargs) -> None:
+        for v in kwargs.values():
+            assert isinstance(
+                v, (list, np.ndarray, torch.Tensor)
+            ), "MaskData only supports list, numpy arrays, and torch tensors."
+        self._stats = dict(**kwargs)
+
+    def __setitem__(self, key: str, item: Any) -> None:
+        assert isinstance(
+            item, (list, np.ndarray, torch.Tensor)
+        ), "MaskData only supports list, numpy arrays, and torch tensors."
+        self._stats[key] = item
+
+    def __delitem__(self, key: str) -> None:
+        del self._stats[key]
+
+    def __getitem__(self, key: str) -> Any:
+        return self._stats[key]
+
+    def items(self) -> ItemsView[str, Any]:
+        return self._stats.items()
+
+    def filter(self, keep: torch.Tensor) -> None:
+        for k, v in self._stats.items():
+            if v is None:
+                self._stats[k] = None
+            elif isinstance(v, torch.Tensor):
+                self._stats[k] = v[torch.as_tensor(keep, device=v.device)]
+            elif isinstance(v, np.ndarray):
+                self._stats[k] = v[keep.detach().cpu().numpy()]
+            elif isinstance(v, list) and keep.dtype == torch.bool:
+                self._stats[k] = [a for i, a in enumerate(v) if keep[i]]
+            elif isinstance(v, list):
+                self._stats[k] = [v[i] for i in keep]
+            else:
+                raise TypeError(f"MaskData key {k} has an unsupported type {type(v)}.")
+
+    def cat(self, new_stats: "MaskData") -> None:
+        for k, v in new_stats.items():
+            if k not in self._stats or self._stats[k] is None:
+                self._stats[k] = deepcopy(v)
+            elif isinstance(v, torch.Tensor):
+                self._stats[k] = torch.cat([self._stats[k], v], dim=0)
+            elif isinstance(v, np.ndarray):
+                self._stats[k] = np.concatenate([self._stats[k], v], axis=0)
+            elif isinstance(v, list):
+                self._stats[k] = self._stats[k] + deepcopy(v)
+            else:
+                raise TypeError(f"MaskData key {k} has an unsupported type {type(v)}.")
+
+    def to_numpy(self) -> None:
+        for k, v in self._stats.items():
+            if isinstance(v, torch.Tensor):
+                self._stats[k] = v.detach().cpu().numpy()
+
+
+def is_box_near_crop_edge(
+    boxes: torch.Tensor, crop_box: List[int], orig_box: List[int], atol: float = 20.0
+) -> torch.Tensor:
+    """Filter masks at the edge of a crop, but not at the edge of the original image."""
+    crop_box_torch = torch.as_tensor(crop_box, dtype=torch.float, device=boxes.device)
+    orig_box_torch = torch.as_tensor(orig_box, dtype=torch.float, device=boxes.device)
+    boxes = uncrop_boxes_xyxy(boxes, crop_box).float()
+    near_crop_edge = torch.isclose(boxes, crop_box_torch[None, :], atol=atol, rtol=0)
+    near_image_edge = torch.isclose(boxes, orig_box_torch[None, :], atol=atol, rtol=0)
+    near_crop_edge = torch.logical_and(near_crop_edge, ~near_image_edge)
+    return torch.any(near_crop_edge, dim=1)
+
+
+def box_xyxy_to_xywh(box_xyxy: torch.Tensor) -> torch.Tensor:
+    box_xywh = deepcopy(box_xyxy)
+    box_xywh[2] = box_xywh[2] - box_xywh[0]
+    box_xywh[3] = box_xywh[3] - box_xywh[1]
+    return box_xywh
+
+
+def batch_iterator(batch_size: int, *args) -> Generator[List[Any], None, None]:
+    assert len(args) > 0 and all(
+        len(a) == len(args[0]) for a in args
+    ), "Batched iteration must have inputs of all the same size."
+    n_batches = len(args[0]) // batch_size + int(len(args[0]) % batch_size != 0)
+    for b in range(n_batches):
+        yield [arg[b * batch_size : (b + 1) * batch_size] for arg in args]
+
+
+def mask_to_rle_pytorch(tensor: torch.Tensor) -> List[Dict[str, Any]]:
+    """
+    Encodes masks to an uncompressed RLE, in the format expected by
+    pycoco tools.
+    """
+    # Put in fortran order and flatten h,w
+    b, h, w = tensor.shape
+    tensor = tensor.permute(0, 2, 1).flatten(1)
+
+    # Compute change indices
+    diff = tensor[:, 1:] ^ tensor[:, :-1]
+    change_indices = diff.nonzero()
+
+    # Encode run length
+    out = []
+    for i in range(b):
+        cur_idxs = change_indices[change_indices[:, 0] == i, 1]
+        cur_idxs = torch.cat(
+            [
+                torch.tensor([0], dtype=cur_idxs.dtype, device=cur_idxs.device),
+                cur_idxs + 1,
+                torch.tensor([h * w], dtype=cur_idxs.dtype, device=cur_idxs.device),
+            ]
+        )
+        btw_idxs = cur_idxs[1:] - cur_idxs[:-1]
+        counts = [] if tensor[i, 0] == 0 else [0]
+        counts.extend(btw_idxs.detach().cpu().tolist())
+        out.append({"size": [h, w], "counts": counts})
+    return out
+
+
+def rle_to_mask(rle: Dict[str, Any]) -> np.ndarray:
+    """Compute a binary mask from an uncompressed RLE."""
+    h, w = rle["size"]
+    mask = np.empty(h * w, dtype=bool)
+    idx = 0
+    parity = False
+    for count in rle["counts"]:
+        mask[idx : idx + count] = parity
+        idx += count
+        parity ^= True
+    mask = mask.reshape(w, h)
+    return mask.transpose()  # Put in C order
+
+
+def area_from_rle(rle: Dict[str, Any]) -> int:
+    return sum(rle["counts"][1::2])
+
+
+def calculate_stability_score(
+    masks: torch.Tensor, mask_threshold: float, threshold_offset: float
+) -> torch.Tensor:
+    """
+    Computes the stability score for a batch of masks. The stability
+    score is the IoU between the binary masks obtained by thresholding
+    the predicted mask logits at high and low values.
+    """
+    # One mask is always contained inside the other.
+    # Save memory by preventing unnecessary cast to torch.int64
+    intersections = (
+        (masks > (mask_threshold + threshold_offset))
+        .sum(-1, dtype=torch.int16)
+        .sum(-1, dtype=torch.int32)
+    )
+    unions = (
+        (masks > (mask_threshold - threshold_offset))
+        .sum(-1, dtype=torch.int16)
+        .sum(-1, dtype=torch.int32)
+    )
+    return intersections / unions
+
+
+def build_point_grid(n_per_side: int) -> np.ndarray:
+    """Generates a 2D grid of points evenly spaced in [0,1]x[0,1]."""
+    offset = 1 / (2 * n_per_side)
+    points_one_side = np.linspace(offset, 1 - offset, n_per_side)
+    points_x = np.tile(points_one_side[None, :], (n_per_side, 1))
+    points_y = np.tile(points_one_side[:, None], (1, n_per_side))
+    points = np.stack([points_x, points_y], axis=-1).reshape(-1, 2)
+    return points
+
+
+def build_all_layer_point_grids(
+    n_per_side: int, n_layers: int, scale_per_layer: int
+) -> List[np.ndarray]:
+    """Generates point grids for all crop layers."""
+    points_by_layer = []
+    for i in range(n_layers + 1):
+        n_points = int(n_per_side / (scale_per_layer**i))
+        points_by_layer.append(build_point_grid(n_points))
+    return points_by_layer
+
+
+def generate_crop_boxes(
+    im_size: Tuple[int, ...], n_layers: int, overlap_ratio: float
+) -> Tuple[List[List[int]], List[int]]:
+    """
+    Generates a list of crop boxes of different sizes. Each layer
+    has (2**i)**2 boxes for the ith layer.
+    """
+    crop_boxes, layer_idxs = [], []
+    im_h, im_w = im_size
+    short_side = min(im_h, im_w)
+
+    # Original image
+    crop_boxes.append([0, 0, im_w, im_h])
+    layer_idxs.append(0)
+
+    def crop_len(orig_len, n_crops, overlap):
+        return int(math.ceil((overlap * (n_crops - 1) + orig_len) / n_crops))
+
+    for i_layer in range(n_layers):
+        n_crops_per_side = 2 ** (i_layer + 1)
+        overlap = int(overlap_ratio * short_side * (2 / n_crops_per_side))
+
+        crop_w = crop_len(im_w, n_crops_per_side, overlap)
+        crop_h = crop_len(im_h, n_crops_per_side, overlap)
+
+        crop_box_x0 = [int((crop_w - overlap) * i) for i in range(n_crops_per_side)]
+        crop_box_y0 = [int((crop_h - overlap) * i) for i in range(n_crops_per_side)]
+
+        # Crops in XYWH format
+        for x0, y0 in product(crop_box_x0, crop_box_y0):
+            box = [x0, y0, min(x0 + crop_w, im_w), min(y0 + crop_h, im_h)]
+            crop_boxes.append(box)
+            layer_idxs.append(i_layer + 1)
+
+    return crop_boxes, layer_idxs
+
+
+def uncrop_boxes_xyxy(boxes: torch.Tensor, crop_box: List[int]) -> torch.Tensor:
+    x0, y0, _, _ = crop_box
+    offset = torch.tensor([[x0, y0, x0, y0]], device=boxes.device)
+    # Check if boxes has a channel dimension
+    if len(boxes.shape) == 3:
+        offset = offset.unsqueeze(1)
+    return boxes + offset
+
+
+def uncrop_points(points: torch.Tensor, crop_box: List[int]) -> torch.Tensor:
+    x0, y0, _, _ = crop_box
+    offset = torch.tensor([[x0, y0]], device=points.device)
+    # Check if points has a channel dimension
+    if len(points.shape) == 3:
+        offset = offset.unsqueeze(1)
+    return points + offset
+
+
+def uncrop_masks(
+    masks: torch.Tensor, crop_box: List[int], orig_h: int, orig_w: int
+) -> torch.Tensor:
+    x0, y0, x1, y1 = crop_box
+    if x0 == 0 and y0 == 0 and x1 == orig_w and y1 == orig_h:
+        return masks
+    # Coordinate transform masks
+    pad_x, pad_y = orig_w - (x1 - x0), orig_h - (y1 - y0)
+    pad = (x0, pad_x - x0, y0, pad_y - y0)
+    return torch.nn.functional.pad(masks, pad, value=0)
+
+
+def remove_small_regions(
+    mask: np.ndarray, area_thresh: float, mode: str
+) -> Tuple[np.ndarray, bool]:
+    """
+    Removes small disconnected regions and holes in a mask. Returns the
+    mask and an indicator of if the mask has been modified.
+    """
+    import cv2  # type: ignore
+
+    assert mode in ["holes", "islands"]
+    correct_holes = mode == "holes"
+    working_mask = (correct_holes ^ mask).astype(np.uint8)
+    n_labels, regions, stats, _ = cv2.connectedComponentsWithStats(working_mask, 8)
+    sizes = stats[:, -1][1:]  # Row 0 is background label
+    small_regions = [i + 1 for i, s in enumerate(sizes) if s < area_thresh]
+    if len(small_regions) == 0:
+        return mask, False
+    fill_labels = [0] + small_regions
+    if not correct_holes:
+        fill_labels = [i for i in range(n_labels) if i not in fill_labels]
+        # If every region is below threshold, keep largest
+        if len(fill_labels) == 0:
+            fill_labels = [int(np.argmax(sizes)) + 1]
+    mask = np.isin(regions, fill_labels)
+    return mask, True
+
+
+def coco_encode_rle(uncompressed_rle: Dict[str, Any]) -> Dict[str, Any]:
+    from pycocotools import mask as mask_utils  # type: ignore
+
+    h, w = uncompressed_rle["size"]
+    rle = mask_utils.frPyObjects(uncompressed_rle, h, w)
+    rle["counts"] = rle["counts"].decode("utf-8")  # Necessary to serialize with json
+    return rle
+
+
+def batched_mask_to_box(masks: torch.Tensor) -> torch.Tensor:
+    """
+    Calculates boxes in XYXY format around masks. Return [0,0,0,0] for
+    an empty mask. For input shape C1xC2x...xHxW, the output shape is C1xC2x...x4.
+    """
+    # torch.max below raises an error on empty inputs, just skip in this case
+    if torch.numel(masks) == 0:
+        return torch.zeros(*masks.shape[:-2], 4, device=masks.device)
+
+    # Normalize shape to CxHxW
+    shape = masks.shape
+    h, w = shape[-2:]
+    if len(shape) > 2:
+        masks = masks.flatten(0, -3)
+    else:
+        masks = masks.unsqueeze(0)
+
+    # Get top and bottom edges
+    in_height, _ = torch.max(masks, dim=-1)
+    in_height_coords = in_height * torch.arange(h, device=in_height.device)[None, :]
+    bottom_edges, _ = torch.max(in_height_coords, dim=-1)
+    in_height_coords = in_height_coords + h * (~in_height)
+    top_edges, _ = torch.min(in_height_coords, dim=-1)
+
+    # Get left and right edges
+    in_width, _ = torch.max(masks, dim=-2)
+    in_width_coords = in_width * torch.arange(w, device=in_width.device)[None, :]
+    right_edges, _ = torch.max(in_width_coords, dim=-1)
+    in_width_coords = in_width_coords + w * (~in_width)
+    left_edges, _ = torch.min(in_width_coords, dim=-1)
+
+    # If the mask is empty the right edge will be to the left of the left edge.
+    # Replace these boxes with [0, 0, 0, 0]
+    empty_filter = (right_edges < left_edges) | (bottom_edges < top_edges)
+    out = torch.stack([left_edges, top_edges, right_edges, bottom_edges], dim=-1)
+    out = out * (~empty_filter).unsqueeze(-1)
+
+    # Return to original shape
+    if len(shape) > 2:
+        out = out.reshape(*shape[:-2], 4)
+    else:
+        out = out[0]
+
+    return out

segment_anything/utils/onnx.py

@@ -0,0 +1,144 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+
+from typing import Tuple
+
+from ..modeling import Sam
+from .amg import calculate_stability_score
+
+
+class SamOnnxModel(nn.Module):
+    """
+    This model should not be called directly, but is used in ONNX export.
+    It combines the prompt encoder, mask decoder, and mask postprocessing of Sam,
+    with some functions modified to enable model tracing. Also supports extra
+    options controlling what information. See the ONNX export script for details.
+    """
+
+    def __init__(
+        self,
+        model: Sam,
+        return_single_mask: bool,
+        use_stability_score: bool = False,
+        return_extra_metrics: bool = False,
+    ) -> None:
+        super().__init__()
+        self.mask_decoder = model.mask_decoder
+        self.model = model
+        self.img_size = model.image_encoder.img_size
+        self.return_single_mask = return_single_mask
+        self.use_stability_score = use_stability_score
+        self.stability_score_offset = 1.0
+        self.return_extra_metrics = return_extra_metrics
+
+    @staticmethod
+    def resize_longest_image_size(
+        input_image_size: torch.Tensor, longest_side: int
+    ) -> torch.Tensor:
+        input_image_size = input_image_size.to(torch.float32)
+        scale = longest_side / torch.max(input_image_size)
+        transformed_size = scale * input_image_size
+        transformed_size = torch.floor(transformed_size + 0.5).to(torch.int64)
+        return transformed_size
+
+    def _embed_points(self, point_coords: torch.Tensor, point_labels: torch.Tensor) -> torch.Tensor:
+        point_coords = point_coords + 0.5
+        point_coords = point_coords / self.img_size
+        point_embedding = self.model.prompt_encoder.pe_layer._pe_encoding(point_coords)
+        point_labels = point_labels.unsqueeze(-1).expand_as(point_embedding)
+
+        point_embedding = point_embedding * (point_labels != -1)
+        point_embedding = point_embedding + self.model.prompt_encoder.not_a_point_embed.weight * (
+            point_labels == -1
+        )
+
+        for i in range(self.model.prompt_encoder.num_point_embeddings):
+            point_embedding = point_embedding + self.model.prompt_encoder.point_embeddings[
+                i
+            ].weight * (point_labels == i)
+
+        return point_embedding
+
+    def _embed_masks(self, input_mask: torch.Tensor, has_mask_input: torch.Tensor) -> torch.Tensor:
+        mask_embedding = has_mask_input * self.model.prompt_encoder.mask_downscaling(input_mask)
+        mask_embedding = mask_embedding + (
+            1 - has_mask_input
+        ) * self.model.prompt_encoder.no_mask_embed.weight.reshape(1, -1, 1, 1)
+        return mask_embedding
+
+    def mask_postprocessing(self, masks: torch.Tensor, orig_im_size: torch.Tensor) -> torch.Tensor:
+        masks = F.interpolate(
+            masks,
+            size=(self.img_size, self.img_size),
+            mode="bilinear",
+            align_corners=False,
+        )
+
+        prepadded_size = self.resize_longest_image_size(orig_im_size, self.img_size).to(torch.int64)
+        masks = masks[..., : prepadded_size[0], : prepadded_size[1]]  # type: ignore
+
+        orig_im_size = orig_im_size.to(torch.int64)
+        h, w = orig_im_size[0], orig_im_size[1]
+        masks = F.interpolate(masks, size=(h, w), mode="bilinear", align_corners=False)
+        return masks
+
+    def select_masks(
+        self, masks: torch.Tensor, iou_preds: torch.Tensor, num_points: int
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        # Determine if we should return the multiclick mask or not from the number of points.
+        # The reweighting is used to avoid control flow.
+        score_reweight = torch.tensor(
+            [[1000] + [0] * (self.model.mask_decoder.num_mask_tokens - 1)]
+        ).to(iou_preds.device)
+        score = iou_preds + (num_points - 2.5) * score_reweight
+        best_idx = torch.argmax(score, dim=1)
+        masks = masks[torch.arange(masks.shape[0]), best_idx, :, :].unsqueeze(1)
+        iou_preds = iou_preds[torch.arange(masks.shape[0]), best_idx].unsqueeze(1)
+
+        return masks, iou_preds
+
+    @torch.no_grad()
+    def forward(
+        self,
+        image_embeddings: torch.Tensor,
+        point_coords: torch.Tensor,
+        point_labels: torch.Tensor,
+        mask_input: torch.Tensor,
+        has_mask_input: torch.Tensor,
+        orig_im_size: torch.Tensor,
+    ):
+        sparse_embedding = self._embed_points(point_coords, point_labels)
+        dense_embedding = self._embed_masks(mask_input, has_mask_input)
+
+        masks, scores = self.model.mask_decoder.predict_masks(
+            image_embeddings=image_embeddings,
+            image_pe=self.model.prompt_encoder.get_dense_pe(),
+            sparse_prompt_embeddings=sparse_embedding,
+            dense_prompt_embeddings=dense_embedding,
+        )
+
+        if self.use_stability_score:
+            scores = calculate_stability_score(
+                masks, self.model.mask_threshold, self.stability_score_offset
+            )
+
+        if self.return_single_mask:
+            masks, scores = self.select_masks(masks, scores, point_coords.shape[1])
+
+        upscaled_masks = self.mask_postprocessing(masks, orig_im_size)
+
+        if self.return_extra_metrics:
+            stability_scores = calculate_stability_score(
+                upscaled_masks, self.model.mask_threshold, self.stability_score_offset
+            )
+            areas = (upscaled_masks > self.model.mask_threshold).sum(-1).sum(-1)
+            return upscaled_masks, scores, stability_scores, areas, masks
+
+        return upscaled_masks, scores, masks