initial commit

NYUv2.py

@@ -0,0 +1,106 @@
+"""NYU Depth Dataset V2"""
+
+import numpy as np
+import h5py
+import datasets
+from datasets import BuilderConfig, Features, Value, SplitGenerator, Split, Array2D, Image
+import hashlib
+
+
+_CITATION = """\
+@inproceedings{Silberman:ECCV12,
+  author    = {Nathan Silberman, Derek Hoiem, Pushmeet Kohli and Rob Fergus},
+  title     = {Indoor Segmentation and Support Inference from RGBD Images},
+  booktitle = {Proceedings of the European Conference on Computer Vision},
+  year      = {2012}
+}
+"""
+
+_DESCRIPTION = """\
+The NYU-Depth V2 data set is comprised of video sequences from a variety of indoor scenes as recorded by both the RGB and Depth cameras from the Microsoft Kinect. It features:
+
+* 1449 densely labeled pairs of aligned RGB and depth images
+* 464 new scenes taken from 3 cities
+* 407,024 new unlabeled frames
+* Each object is labeled with a class and an instance number (cup1, cup2, cup3, etc)
+
+This dataset is useful for various computer vision tasks, including depth estimation, semantic segmentation, and instance segmentation.
+
+This Hugging Face version of the dataset is unofficial. It downloads the data from the original source and converts it to the Hugging Face format.
+"""
+
+_HOMEPAGE = "https://cs.nyu.edu/~fergus/datasets/nyu_depth_v2.html"
+
+_LICENSE = "Unknown"
+
+_URL = "http://horatio.cs.nyu.edu/mit/silberman/nyu_depth_v2/nyu_depth_v2_labeled.mat"
+_FILE_HASH = "520609c519fba3ba5ac58c8fefcc3530"
+
+class NYUv2(datasets.GeneratorBasedBuilder):
+    """NYU Depth Dataset V2"""
+
+    VERSION = datasets.Version("1.0.0")
+
+    BUILDER_CONFIGS = [
+        BuilderConfig(name="default", version=VERSION, description="Default configuration for NYUv2 dataset"),
+    ]
+
+    DEFAULT_CONFIG_NAME = "default"
+
+    def _info(self):
+        features = Features({
+            "image": Image(decode=True),
+            "depth": Array2D(dtype="float32", shape=(640, 480)),
+            "label": Array2D(dtype="int32", shape=(640, 480)),
+            "scene": Value("string"),
+            "scene_type": Value("string"),
+        })
+
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_path = dl_manager.download(_URL)
+
+        # Verify file hash
+        with open(data_path, 'rb') as f:
+            file_hash = hashlib.md5(f.read()).hexdigest()
+        if file_hash != _FILE_HASH:
+            raise ValueError(
+                f"Downloaded file hash '{file_hash}' does not match expected hash '{_FILE_HASH}'. "
+                "The downloaded dataset file might be corrupted or modified."
+            )
+
+        return [
+            SplitGenerator(
+                name="train",
+                gen_kwargs={
+                    "filepath": data_path,
+                    "split": "train",
+                },
+            ),
+        ]
+
+    def _generate_examples(self, filepath, split):
+        with h5py.File(filepath, 'r') as f:
+            images = np.array(f['images'])
+            depths = np.array(f['depths'])
+            instances = np.array(f['instances'])
+            labels = np.array(f['labels'])
+            scenes = [''.join(chr(int(x)) for x in f[y]) for y in f.get('scenes')[0]]
+            scene_types = [''.join(chr(int(x)) for x in f[y]) for y in f.get('sceneTypes')[0]]
+
+        for idx in range(images.shape[0]):
+            yield idx, {
+                "image": images[idx].transpose(1, 2, 0),
+                "depth": depths[idx],
+                "instance": instances[idx],
+                "label": labels[idx],
+                "scene": scenes[idx],
+                "scene_type": scene_types[idx],
+            }

README.md

@@ -0,0 +1,109 @@
+---
+size_categories:
+- 1k<n<10k
+paperswithcode_id: nyuv2
+tags:
+- depth-estimation
+- semantic-segmentation
+dataset_info:
+  features:
+  - name: image
+    dtype: image
+  - name: depth
+    dtype:
+      array2_d:
+        shape:
+        - 640
+        - 480
+        dtype: float32
+  - name: label
+    dtype:
+      array2_d:
+        shape:
+        - 640
+        - 480
+        dtype: int32
+  - name: scene
+    dtype: string
+  - name: scene_type
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 4096466619
+    num_examples: 1449
+  download_size: 2972037809
+  dataset_size: 4096466619
+---
+
+# NYU Depth Dataset V2
+
+This is an unofficial Hugging Face downloading script of the [NYU Depth Dataset V2](https://cs.nyu.edu/~fergus/datasets/nyu_depth_v2.html). It downloads the data from the original source and converts it to the Hugging Face format.
+
+This dataset contains the 1449 densely labeled pairs of aligned RGB and depth images.
+
+- **Homepage:** [NYU Depth Dataset V2](https://cs.nyu.edu/~fergus/datasets/nyu_depth_v2.html)
+- **Paper:** [Indoor Segmentation and Support Inference from RGBD Images](https://cs.nyu.edu/~fergus/datasets/indoor_seg_support.pdf)
+
+
+## Official Description
+
+The NYU-Depth V2 data set is comprised of video sequences from a variety of indoor scenes as recorded by both the RGB and Depth cameras from the Microsoft Kinect. It features:
+
+* 1449 densely labeled pairs of aligned RGB and depth images
+* 464 new scenes taken from 3 cities
+* 407,024 new unlabeled frames
+* Each object is labeled with a class and an instance number (cup1, cup2, cup3, etc)
+
+This dataset is useful for various computer vision tasks, including depth estimation, semantic segmentation, and instance segmentation.
+
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+# Load the dataset
+dataset = load_dataset("0jl/NYUv2", trust_remote_code=True, split="train")
+```
+
+
+### Common Errors
+
+#### `fsspec.exceptions.FSTimeoutError`
+
+Can occur for `datasets==3.0` when the download takes more than 5 minutes.
+
+```python
+import datasets, aiohttp
+dataset = datasets.load_dataset(
+    "0jl/NYUv2",
+    trust_remote_code=True,
+    split="train",
+    storage_options={'client_kwargs': {'timeout': aiohttp.ClientTimeout(total=3600)}}
+)
+```
+
+
+## Dataset Structure
+
+The dataset contains only one training split with the following features:
+
+- `image`: RGB image (PIL.Image.Image, shape: (640, 480, 3))
+- `depth`: Depth map (numpy.ndarray, shape: (640, 480), dtype: float32)
+- `label`: Semantic segmentation labels (numpy.ndarray, shape: (640, 480), dtype: int32)
+- `scene`: Scene name (string)
+- `scene_type`: Scene type (string)
+
+
+### Citation Information
+
+If you use this dataset, please cite the original paper:
+
+```bibtex
+@inproceedings{Silberman:ECCV12,
+  author    = {Nathan Silberman, Derek Hoiem, Pushmeet Kohli and Rob Fergus},
+  title     = {Indoor Segmentation and Support Inference from RGBD Images},
+  booktitle = {Proceedings of the European Conference on Computer Vision},
+  year      = {2012}
+}
+```