occluded-trellis

app.py

@@ -1,20 +1,19 @@
-import gradio as gr
-import spaces
-from gradio_litmodel3d import LitModel3D
-
 import os
 import shutil
-os.environ['SPCONV_ALGO'] = 'native'
 from typing import *
-import torch
-import numpy as np
+
+import gradio as gr
 import imageio
-from easydict import EasyDict as edict
+import numpy as np
+import torch
 from PIL import Image
+from easydict import EasyDict as edict
+from gradio_litmodel3d import LitModel3D
 from trellis.pipelines import TrellisImageTo3DPipeline
 from trellis.representations import Gaussian, MeshExtractResult
 from trellis.utils import render_utils, postprocessing_utils
 
+from occluded_trellis import apply_occluded_patch
 
 MAX_SEED = np.iinfo(np.int32).max
 TMP_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'tmp')
@@ -24,42 +23,13 @@ os.makedirs(TMP_DIR, exist_ok=True)
 def start_session(req: gr.Request):
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     os.makedirs(user_dir, exist_ok=True)
-    
-    
+
+
 def end_session(req: gr.Request):
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     shutil.rmtree(user_dir)
 
 
-def preprocess_image(image: Image.Image) -> Image.Image:
-    """
-    Preprocess the input image.
-
-    Args:
-        image (Image.Image): The input image.
-
-    Returns:
-        Image.Image: The preprocessed image.
-    """
-    processed_image = pipeline.preprocess_image(image)
-    return processed_image
-
-
-def preprocess_images(images: List[Tuple[Image.Image, str]]) -> List[Image.Image]:
-    """
-    Preprocess a list of input images.
-    
-    Args:
-        images (List[Tuple[Image.Image, str]]): The input images.
-        
-    Returns:
-        List[Image.Image]: The preprocessed images.
-    """
-    images = [image[0] for image in images]
-    processed_images = [pipeline.preprocess_image(image) for image in images]
-    return processed_images
-
-
 def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
     return {
         'gaussian': {
@@ -75,8 +45,8 @@ def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
             'faces': mesh.faces.cpu().numpy(),
         },
     }
-    
-    
+
+
 def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
     gs = Gaussian(
         aabb=state['gaussian']['aabb'],
@@ -91,12 +61,12 @@ def unpack_state(state: dict) -> Tuple[Gaussian, edict, str]:
     gs._scaling = torch.tensor(state['gaussian']['_scaling'], device='cuda')
     gs._rotation = torch.tensor(state['gaussian']['_rotation'], device='cuda')
     gs._opacity = torch.tensor(state['gaussian']['_opacity'], device='cuda')
-    
+
     mesh = edict(
         vertices=torch.tensor(state['mesh']['vertices'], device='cuda'),
         faces=torch.tensor(state['mesh']['faces'], device='cuda'),
     )
-    
+
     return gs, mesh
 
 
@@ -107,69 +77,52 @@ def get_seed(randomize_seed: bool, seed: int) -> int:
     return np.random.randint(0, MAX_SEED) if randomize_seed else seed
 
 
-@spaces.GPU
 def image_to_3d(
-    image: Image.Image,
-    multiimages: List[Tuple[Image.Image, str]],
-    is_multiimage: bool,
-    seed: int,
-    ss_guidance_strength: float,
-    ss_sampling_steps: int,
-    slat_guidance_strength: float,
-    slat_sampling_steps: int,
-    multiimage_algo: Literal["multidiffusion", "stochastic"],
-    req: gr.Request,
+        image: Image.Image,
+        mask: Image.Image,
+        seed: int,
+        ss_guidance_strength: float,
+        ss_sampling_steps: int,
+        slat_guidance_strength: float,
+        slat_sampling_steps: int,
+        req: gr.Request,
 ) -> Tuple[dict, str]:
     """
     Convert an image to a 3D model.
 
     Args:
         image (Image.Image): The input image.
-        multiimages (List[Tuple[Image.Image, str]]): The input images in multi-image mode.
-        is_multiimage (bool): Whether is in multi-image mode.
         seed (int): The random seed.
         ss_guidance_strength (float): The guidance strength for sparse structure generation.
         ss_sampling_steps (int): The number of sampling steps for sparse structure generation.
         slat_guidance_strength (float): The guidance strength for structured latent generation.
         slat_sampling_steps (int): The number of sampling steps for structured latent generation.
-        multiimage_algo (Literal["multidiffusion", "stochastic"]): The algorithm for multi-image generation.
 
     Returns:
         dict: The information of the generated 3D model.
         str: The path to the video of the 3D model.
     """
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
-    if not is_multiimage:
-        outputs = pipeline.run(
-            image,
-            seed=seed,
-            formats=["gaussian", "mesh"],
-            preprocess_image=False,
-            sparse_structure_sampler_params={
-                "steps": ss_sampling_steps,
-                "cfg_strength": ss_guidance_strength,
-            },
-            slat_sampler_params={
-                "steps": slat_sampling_steps,
-                "cfg_strength": slat_guidance_strength,
-            },
-        )
-    else:
-        outputs = pipeline.run_multi_image(
-            [image[0] for image in multiimages],
-            seed=seed,
-            formats=["gaussian", "mesh"],
-            preprocess_image=False,
-            sparse_structure_sampler_params={
-                "steps": ss_sampling_steps,
-                "cfg_strength": ss_guidance_strength,
-            },
-            slat_sampler_params={
-                "steps": slat_sampling_steps,
-                "cfg_strength": slat_guidance_strength,
-            },
-            mode=multiimage_algo,
-        )
+    # if image.mode == "RGBA":
+    #     image_np = np.array(image)
+    #     alpha = image_np[..., 3]
+    #     if np.all(alpha == 255):
+    #         image = image.convert("RGB")
+    outputs = pipeline.run(
+        image,
+        mask,
+        seed=seed,
+        formats=["gaussian", "mesh"],
+        preprocess_image=True,
+        sparse_structure_sampler_params={
+            "steps": ss_sampling_steps,
+            "cfg_strength": ss_guidance_strength,
+        },
+        slat_sampler_params={
+            "steps": slat_sampling_steps,
+            "cfg_strength": slat_guidance_strength,
+        },
+    )
     video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
     video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
     video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
@@ -180,12 +133,11 @@ def image_to_3d(
     return state, video_path
 
 
-@spaces.GPU(duration=90)
 def extract_glb(
-    state: dict,
-    mesh_simplify: float,
-    texture_size: int,
-    req: gr.Request,
+        state: dict,
+        mesh_simplify: float,
+        texture_size: int,
+        req: gr.Request,
 ) -> Tuple[str, str]:
     """
     Extract a GLB file from the 3D model.
@@ -207,7 +159,6 @@ def extract_glb(
     return glb_path, glb_path
 
 
-@spaces.GPU
 def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
     """
     Extract a Gaussian file from the 3D model.
@@ -226,57 +177,39 @@ def extract_gaussian(state: dict, req: gr.Request) -> Tuple[str, str]:
     return gaussian_path, gaussian_path
 
 
-def prepare_multi_example() -> List[Image.Image]:
-    multi_case = list(set([i.split('_')[0] for i in os.listdir("assets/example_multi_image")]))
-    images = []
-    for case in multi_case:
-        _images = []
-        for i in range(1, 4):
-            img = Image.open(f'assets/example_multi_image/{case}_{i}.png')
-            W, H = img.size
-            img = img.resize((int(W / H * 512), 512))
-            _images.append(np.array(img))
-        images.append(Image.fromarray(np.concatenate(_images, axis=1)))
-    return images
+def extract_mask(editor_content):
+    bg = editor_content.get("background", None)
+    layers = editor_content.get("layers", [])
 
+    if bg is None and not layers:
+        raise ValueError("No background or layers provided")
 
-def split_image(image: Image.Image) -> List[Image.Image]:
-    """
-    Split an image into multiple views.
-    """
-    image = np.array(image)
-    alpha = image[..., 3]
-    alpha = np.any(alpha>0, axis=0)
-    start_pos = np.where(~alpha[:-1] & alpha[1:])[0].tolist()
-    end_pos = np.where(alpha[:-1] & ~alpha[1:])[0].tolist()
-    images = []
-    for s, e in zip(start_pos, end_pos):
-        images.append(Image.fromarray(image[:, s:e+1]))
-    return [preprocess_image(image) for image in images]
+    mask = layers[0]
+    mask = np.array(mask)
+    mask = np.where(mask[..., 3] != 0, 0, 255).astype(
+        np.uint8)
+    mask = np.stack([mask] * 3, -1)
+    mask = Image.fromarray(mask)
+    return [bg, mask]
 
 
 with gr.Blocks(delete_cache=(600, 600)) as demo:
     gr.Markdown("""
-    ## Image to 3D Asset with [TRELLIS](https://trellis3d.github.io/)
-    * Upload an image and click "Generate" to create a 3D asset. If the image has alpha channel, it be used as the mask. Otherwise, we use `rembg` to remove the background.
+    ## Image to 3D Asset with [TRELLIS](https://trellis3d.github.io/), [Occluded version](https://github.com/cortwave/OccludedTrellis)
+    * Upload an image, draw a mask and click "Generate" to create a 3D asset. If the image has alpha channel, it be used as the mask. Otherwise, we use `rembg` to remove the background.
     * If you find the generated 3D asset satisfactory, click "Extract GLB" to extract the GLB file and download it.
-    
-    ✨New: 1) Experimental multi-image support. 2) Gaussian file extraction.
+    * If image is already corrupted even before masking it's highly recommended to upload with your own alpha mask (remove background)
     """)
-    
+
     with gr.Row():
         with gr.Column():
-            with gr.Tabs() as input_tabs:
-                with gr.Tab(label="Single Image", id=0) as single_image_input_tab:
-                    image_prompt = gr.Image(label="Image Prompt", format="png", image_mode="RGBA", type="pil", height=300)
-                with gr.Tab(label="Multiple Images", id=1) as multiimage_input_tab:
-                    multiimage_prompt = gr.Gallery(label="Image Prompt", format="png", type="pil", height=300, columns=3)
-                    gr.Markdown("""
-                        Input different views of the object in separate images. 
-                        
-                        *NOTE: this is an experimental algorithm without training a specialized model. It may not produce the best results for all images, especially those having different poses or inconsistent details.*
-                    """)
-            
+            im_ed = gr.ImageMask(label="Image Prompt",
+                                 format="png",
+                                 image_mode="RGBA",
+                                 type="pil",
+                                 sources=["upload", "clipboard"])
+            image_prompt = gr.Image(visible=False, type="pil", image_mode="RGBA", format="png")
+            mask = gr.Image(visible=False, type="pil")
             with gr.Accordion(label="Generation Settings", open=False):
                 seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
                 randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
@@ -288,14 +221,13 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
                 with gr.Row():
                     slat_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=3.0, step=0.1)
                     slat_sampling_steps = gr.Slider(1, 50, label="Sampling Steps", value=12, step=1)
-                multiimage_algo = gr.Radio(["stochastic", "multidiffusion"], label="Multi-image Algorithm", value="stochastic")
 
             generate_btn = gr.Button("Generate")
-            
+
             with gr.Accordion(label="GLB Extraction Settings", open=False):
                 mesh_simplify = gr.Slider(0.9, 0.98, label="Simplify", value=0.95, step=0.01)
                 texture_size = gr.Slider(512, 2048, label="Texture Size", value=1024, step=512)
-            
+
             with gr.Row():
                 extract_glb_btn = gr.Button("Extract GLB", interactive=False)
                 extract_gs_btn = gr.Button("Extract Gaussian", interactive=False)
@@ -306,68 +238,29 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
         with gr.Column():
             video_output = gr.Video(label="Generated 3D Asset", autoplay=True, loop=True, height=300)
             model_output = LitModel3D(label="Extracted GLB/Gaussian", exposure=10.0, height=300)
-            
+
             with gr.Row():
                 download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
-                download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)  
-    
-    is_multiimage = gr.State(False)
-    output_buf = gr.State()
+                download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
 
-    # Example images at the bottom of the page
-    with gr.Row() as single_image_example:
-        examples = gr.Examples(
-            examples=[
-                f'assets/example_image/{image}'
-                for image in os.listdir("assets/example_image")
-            ],
-            inputs=[image_prompt],
-            fn=preprocess_image,
-            outputs=[image_prompt],
-            run_on_click=True,
-            examples_per_page=64,
-        )
-    with gr.Row(visible=False) as multiimage_example:
-        examples_multi = gr.Examples(
-            examples=prepare_multi_example(),
-            inputs=[image_prompt],
-            fn=split_image,
-            outputs=[multiimage_prompt],
-            run_on_click=True,
-            examples_per_page=8,
-        )
+    output_buf = gr.State()
 
     # Handlers
     demo.load(start_session)
     demo.unload(end_session)
-    
-    single_image_input_tab.select(
-        lambda: tuple([False, gr.Row.update(visible=True), gr.Row.update(visible=False)]),
-        outputs=[is_multiimage, single_image_example, multiimage_example]
-    )
-    multiimage_input_tab.select(
-        lambda: tuple([True, gr.Row.update(visible=False), gr.Row.update(visible=True)]),
-        outputs=[is_multiimage, single_image_example, multiimage_example]
-    )
-    
-    image_prompt.upload(
-        preprocess_image,
-        inputs=[image_prompt],
-        outputs=[image_prompt],
-    )
-    multiimage_prompt.upload(
-        preprocess_images,
-        inputs=[multiimage_prompt],
-        outputs=[multiimage_prompt],
-    )
 
     generate_btn.click(
         get_seed,
         inputs=[randomize_seed, seed],
         outputs=[seed],
+    ).then(
+        extract_mask,
+        inputs=im_ed,
+        outputs=[image_prompt, mask]
     ).then(
         image_to_3d,
-        inputs=[image_prompt, multiimage_prompt, is_multiimage, seed, ss_guidance_strength, ss_sampling_steps, slat_guidance_strength, slat_sampling_steps, multiimage_algo],
+        inputs=[image_prompt, mask, seed, ss_guidance_strength, ss_sampling_steps,
+                slat_guidance_strength, slat_sampling_steps],
         outputs=[output_buf, video_output],
     ).then(
         lambda: tuple([gr.Button(interactive=True), gr.Button(interactive=True)]),
@@ -387,7 +280,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
         lambda: gr.Button(interactive=True),
         outputs=[download_glb],
     )
-    
+
     extract_gs_btn.click(
         extract_gaussian,
         inputs=[output_buf],
@@ -401,14 +294,10 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
         lambda: gr.Button(interactive=False),
         outputs=[download_glb],
     )
-    
 
 # Launch the Gradio app
 if __name__ == "__main__":
     pipeline = TrellisImageTo3DPipeline.from_pretrained("JeffreyXiang/TRELLIS-image-large")
+    pipeline = apply_occluded_patch(pipeline)
     pipeline.cuda()
-    try:
-        pipeline.preprocess_image(Image.fromarray(np.zeros((512, 512, 3), dtype=np.uint8)))    # Preload rembg
-    except:
-        pass
     demo.launch()

requirements.txt

@@ -23,4 +23,5 @@ transformers==4.46.3
 gradio_litmodel3d==0.0.1
 https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.0.post2/flash_attn-2.7.0.post2+cu12torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
 https://huggingface.co/spaces/JeffreyXiang/TRELLIS/resolve/main/wheels/diff_gaussian_rasterization-0.0.0-cp310-cp310-linux_x86_64.whl?download=true
-https://huggingface.co/spaces/JeffreyXiang/TRELLIS/resolve/main/wheels/nvdiffrast-0.3.3-cp310-cp310-linux_x86_64.whl?download=true
+https://huggingface.co/spaces/JeffreyXiang/TRELLIS/resolve/main/wheels/nvdiffrast-0.3.3-cp310-cp310-linux_x86_64.whl?download=true
+occluded-trellis=0.1.1