Upload README.md with huggingface_hub

README.md

@@ -1,4 +1,27 @@
 ---
+annotations_creators:
+- expert-generated
+language:
+- en
+license: apple-ascl
+multilinguality: monolingual
+pretty_name: Apple Dense Material Segmentation (DMS) – Stratified Split
+size_categories:
+- 10K<n<100K
+source_datasets:
+- extended|open-images-v7
+tags:
+- material-segmentation
+- semantic-segmentation
+- dense-prediction
+- materials
+- segformer
+- mask2former
+- stratified-split
+task_categories:
+- image-segmentation
+task_ids:
+- semantic-segmentation
 dataset_info:
   features:
   - name: image
@@ -9,23 +32,172 @@ dataset_info:
     dtype: string
   splits:
   - name: train
-    num_bytes: 8973840321
     num_examples: 33118
   - name: validation
-    num_bytes: 1205036342
     num_examples: 4138
   - name: test
-    num_bytes: 1174346942
     num_examples: 4140
-  download_size: 11730154334
-  dataset_size: 11353223605
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
-  - split: test
-    path: data/test-*
 ---
+
+# Apple Dense Material Segmentation (DMS) – Stratified 80/10/10 Split
+
+A **pixel-level material segmentation** dataset containing ~41K images with dense annotations across **57 material categories**. Originally released by Apple as part of the [Dense Material Segmentation (DMS)](https://machinelearning.apple.com/research/dense-material-segmentation) research project.
+
+This version uses a **custom stratified 80/10/10 split** (vs Apple's original 54/23/23) to maximise training data while maintaining representative validation and test sets.
+
+## Why a Custom Split?
+
+Apple's original split reserves nearly half the data for evaluation (23% val + 23% test). Our re-split allocates **80% to training** while using **stratified sampling** (based on the dominant material class per image) to keep val/test distributions aligned with the training set.
+
+### Split Quality Comparison
+
+| Metric | Original (Apple) | Custom (Stratified) | Improvement |
+|--------|------------------|---------------------|-------------|
+| Train size | 22,492 (54%) | **33,118 (80%)** | +47% more training data |
+| JSD train↔val | 0.0524 | **0.0158** | ✅ **70% lower divergence** |
+| JSD train↔test | 0.0526 | **0.0163** | ✅ **69% lower divergence** |
+| Classes in all splits | 53/57 | 53/57 | Equal coverage |
+
+> **JSD** = Jensen-Shannon Divergence between pixel-level class distributions. Lower values mean the evaluation sets better represent the training distribution, leading to more reliable metrics.
+
+## Dataset Description
+
+Each sample consists of:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `image` | `PIL.Image` | RGB input image |
+| `label` | `PIL.Image` | Single-channel segmentation mask (pixel values = class indices 0–56) |
+| `image_id` | `string` | Unique image identifier |
+
+### Splits
+
+| Split | Samples | Percentage |
+|-------|---------|------------|
+| Train | 33,118 | 80.0% |
+| Validation | 4,138 | 10.0% |
+| Test | 4,140 | 10.0% |
+| **Total** | **41,396** | **100%** |
+
+### Material Classes (57)
+
+<details>
+<summary>Click to expand full class list</summary>
+
+| ID | Material | ID | Material | ID | Material |
+|----|----------|----|----------|----|----------|
+| 0 | No label | 19 | Gemstone/quartz | 38 | Sky |
+| 1 | Animal skin | 20 | Glass | 39 | Snow |
+| 2 | Bone/teeth/horn | 21 | Hair | 40 | Soap |
+| 3 | Brickwork | 22 | I cannot tell | 41 | Soil/mud |
+| 4 | Cardboard | 23 | Ice | 42 | Sponge |
+| 5 | Carpet/rug | 24 | Leather | 43 | Stone, natural |
+| 6 | Ceiling tile | 25 | Liquid, non-water | 44 | Stone, polished |
+| 7 | Ceramic | 26 | Metal | 45 | Styrofoam |
+| 8 | Chalkboard/blackboard | 27 | Mirror | 46 | Tile |
+| 9 | Clutter | 28 | Not on list | 47 | Wallpaper |
+| 10 | Concrete | 29 | Paint/plaster/enamel | 48 | Water |
+| 11 | Cork/corkboard | 30 | Paper | 49 | Wax |
+| 12 | Engineered stone | 31 | Pearl | 50 | Whiteboard |
+| 13 | Fabric/cloth | 32 | Photograph/painting | 51 | Wicker |
+| 14 | Fiberglass wool | 33 | Plastic, clear | 52 | Wood |
+| 15 | Fire | 34 | Plastic, non-clear | 53 | Wood, tree |
+| 16 | Foliage | 35 | Rubber/latex | 54 | Bad polygon |
+| 17 | Food | 36 | Sand | 55 | Multiple materials |
+| 18 | Fur | 37 | Skin/lips | 56 | Asphalt |
+
+</details>
+
+## Usage
+
+### Loading the Dataset
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("AllanK24/apple-dms-materials-v2")
+
+# Access splits
+train_ds = dataset["train"]      # 33,118 samples
+val_ds = dataset["validation"]   #  4,138 samples
+test_ds = dataset["test"]        #  4,140 samples
+
+# View a sample
+sample = train_ds[0]
+sample["image"].show()    # RGB image
+sample["label"].show()    # Segmentation mask
+```
+
+### Training with SegFormer / Mask2Former
+
+```python
+from transformers import SegformerForSemanticSegmentation, SegformerImageProcessorFast
+import json
+from huggingface_hub import hf_hub_download
+
+# Load class info
+class_info_path = hf_hub_download(
+    repo_id="AllanK24/apple-dms-materials-v2",
+    filename="class_info.json",
+    repo_type="dataset",
+)
+with open(class_info_path) as f:
+    class_info = json.load(f)
+
+id2label = {int(k): v for k, v in class_info["id2label"].items()}
+label2id = class_info["label2id"]
+num_labels = class_info["num_labels"]
+
+# Initialize model
+model = SegformerForSemanticSegmentation.from_pretrained(
+    "nvidia/segformer-b2-finetuned-ade-512-512",
+    num_labels=num_labels,
+    id2label=id2label,
+    label2id=label2id,
+    ignore_mismatched_sizes=True,
+)
+
+# Initialize processor
+processor = SegformerImageProcessorFast.from_pretrained(
+    "nvidia/segformer-b2-finetuned-ade-512-512"
+)
+
+# Apply transforms
+def transforms(batch):
+    images = [x.convert("RGB") for x in batch["image"]]
+    labels = [x for x in batch["label"]]
+    return processor(images=images, segmentation_maps=labels, return_tensors="pt")
+
+train_ds.set_transform(transforms)
+```
+
+## Stratification Method
+
+The split was created using a two-level stratified sampling approach:
+1. **Dominant class extraction** – For each image, the material class with the most pixels (excluding "No label") is identified.
+2. **First split** – Images are stratified into 80% train vs 20% eval using `StratifiedShuffleSplit`.
+3. **Second split** – The 20% eval pool is stratified 50/50 into validation and test.
+4. **Rare class handling** – Classes with <5 total images go directly to train; classes with <2 images in the eval pool are randomly assigned between val/test.
+
+Seed: `42` (for reproducibility).
+
+## Source & Preparation
+
+- **Original dataset**: [Apple DMS](https://github.com/apple/ml-dms-dataset) with images from [Open Images V7](https://storage.googleapis.com/openimages/web/index.html)
+- **Original split** (v1): [AllanK24/apple-dms-materials](https://huggingface.co/datasets/AllanK24/apple-dms-materials)
+- **Preparation pipeline**: Download → resize/align (`prepare_images.py`) → validate (`check_images.py`, 41,385/41,396 passed) → stratified re-split
+
+## Citation
+
+```bibtex
+@article{upchurch2022dense,
+  title={Dense Material Segmentation with Context-Aware Network},
+  author={Upchurch, Paul and Niu, Ransen},
+  year={2022},
+  url={https://machinelearning.apple.com/research/dense-material-segmentation}
+}
+```
+
+## License
+
+Released under the [Apple Sample Code License (ASCL)](https://developer.apple.com/sample-code/license/apple-sample-code-license/). Source images are from Open Images V7 (primarily CC BY 2.0). See the [original repository](https://github.com/apple/ml-dms-dataset) for full licensing details.