AllanK24/apple-dms-materials

Apple Dense Material Segmentation (DMS) Dataset

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) research project.

Note: This is a mirror prepared for direct use with the HuggingFace 🤗 datasets library. The source images originate from Open Images V7, and material annotations were created by Apple. Some images (~6%) from the original dataset could not be retrieved from Open Images and are therefore absent.

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	22,492	54.3%
Validation	9,412	22.7%
Test	9,492	22.9%
Total	41,396	100%

The split assignments follow the original Apple DMS partition.

Material Classes (57)

Click to expand full class list

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

Usage

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("AllanK24/apple-dms-materials")

# Access splits
train_ds = dataset["train"]
val_ds = dataset["validation"]
test_ds = dataset["test"]

# View a sample
sample = train_ds[0]
print(sample["image_id"])   # e.g. "22491"
sample["image"].show()      # RGB image
sample["label"].show()      # Segmentation mask

Training with SegFormer

from transformers import SegformerForSemanticSegmentation, SegformerImageProcessorFast
import json

# Load class info
from huggingface_hub import hf_hub_download
class_info_path = hf_hub_download(
    repo_id="AllanK24/apple-dms-materials",
    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"
)

Applying Transforms

def transforms(batch):
    images = [x.convert("RGB") for x in batch["image"]]
    labels = [x for x in batch["label"]]
    inputs = processor(images=images, segmentation_maps=labels, return_tensors="pt")
    return inputs

train_ds.set_transform(transforms)

Dataset Preparation

This dataset was prepared from the original Apple DMS release using the following pipeline:

1. Download – Source images retrieved from Open Images V7 using URLs in Apple's metadata.
2. Resize & align – Images resized to match label dimensions using Apple's prepare_images.py.
3. Validation – Image–label consistency verified with Apple's check_images.py (41,385 / 41,396 passed; 11 minor rotation warnings).

Citation

If you use this dataset, please cite the original Apple paper:

@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

This dataset is released under the Apple Sample Code License (ASCL). The source images are from Open Images V7 and are subject to their respective licenses (primarily CC BY 2.0). Please refer to the original repository for full licensing details.