TextureADE

Real scenes carrying several appearance transitions each, mined from the ADE20K validation split.

One of the four evaluation routes in the ICLR 2027 submission on sub-semantic image segmentation: partitioning an image into regions that are coherent in appearance and describable in language, but that need not correspond to any object, part or material class.

• Images: 212
• Code: github.com/aviadcohz/Qwen2SAM_Detecture_Benchmark
• Weights: aviadcohz/Detecture-ICLR-2027
• All four routes in one download: aviadcohz/Detecture_ICLR_Benchmarking

Layout

ADE20k_Detecture/
├── images/            RGB images
├── textures_mask/     per-texture binary masks, <id>_mask_<k>.png
├── metadata.json      image paths, mask paths, descriptions
└── summary.json       dataset statistics

The three real-world routes also carry masks/ and overlays/; overlays are visualisations, not ground truth.

The directory inside this repo is named ADE20k_Detecture rather than TextureADE, because that is the name the evaluation configs resolve (fairness_baseline_suite/src/paths.py). Paths inside metadata.json are relative to the repository root, so the folder can be placed anywhere.

Use

cd ~/datasets
git lfs install
git clone https://huggingface.co/datasets/aviadcohz/TextureADE
mv TextureADE/ADE20k_Detecture . && rm -rf TextureADE

Then, from the benchmark repo:

cd Qwen2SAM_Detecture_Benchmark/fairness_baseline_suite
PYTHONPATH=src python src/run_fairness.py --model detecture --dataset TextureADE

Evaluation protocol

Every number reported on this route comes from one protocol applied identically to every method: no ground-truth region count in the prompt, no inverse-mask completion, no truncation of proposals to a known count, and no dropping of images where a method returns nothing. The region count is inferred, never supplied. Results obtained this way are not comparable to evaluations that supply it.

Provenance

Mined from the natural ADE20K validation split by a geometry-first procedure: connected components are merged into at most five candidate regions each covering at least 1% of image area, scored on mask structure and boundary geometry, and admitted only above a fixed threshold. A frozen vision-language annotator is queried afterwards, against a region that has already been accepted, so language never proposes regions.

Licence

CC-BY-4.0 for this packaging. Upstream corpora keep their own terms: ADE20K.