Add dataset card

README.md

@@ -0,0 +1,68 @@
+---
+task_categories:
+  - image-segmentation
+license: cc-by-nc-4.0
+library_name: mmsegmentation
+---
+
+# CorrCLIPv2: Pre-generated Masks and Prebuilt Memory Banks
+
+This repository contains the pre-generated region masks and the prebuilt memory banks used to reproduce **CorrCLIPv2**, the journal extension of [CorrCLIP: Reconstructing Patch Correlations in CLIP for Open-Vocabulary Semantic Segmentation](https://arxiv.org/abs/2411.10086).
+
+CorrCLIPv2 augments single-image open-vocabulary segmentation with cross-image semantic support retrieved from an offline memory bank built purely from unlabeled images — no annotation, no caption, no training. Downloading the artifacts here lets you skip both the mask generation and the bank construction step.
+
+For the code, installation instructions and evaluation scripts, see the GitHub repository:
+[https://github.com/zdk258/CorrCLIP](https://github.com/zdk258/CorrCLIP) (the `CorrCLIPv2/` subdirectory).
+
+## Contents
+
+| File | Size | What it is |
+|---|---|---|
+| `instance_mask.zip` | 110 M | EntitySeg region masks for the ten standard benchmarks (VOC, Context, COCO, ADE20K, ADE847, Cityscapes) |
+| `memory_bank/clip_embeddings/{bank}_metaclip_b.pt` | 2.8 G / 1.4 G | CLIP region features, **CorrCLIPv2-B** main store |
+| `memory_bank/clip_embeddings/{bank}_metaclip_l.pt` | 4.2 G / 2.2 G | CLIP region features, **CorrCLIPv2-L** main store |
+| `memory_bank/clip_embeddings/{bank}_dfnclip_b.pt` | 2.8 G / 1.4 G | DFN auxiliary store for the multi-CLIP ensemble, shared by B and L |
+| `memory_bank/vfm_embeddings/{bank}_radiov3.faiss` | 190 M / 108 M | RADIOv3 retrieval index (IDMap2), shared by B and L |
+| `memory_bank/bank_images_{gpic,monet,sa1b}.txt` | 10 M | Source image list each disjoint bank was built from, one path per line |
+
+Two banks are provided: `coco` (built from COCO train2017, used for the main results) and `monet` (built from MONET Z-Image samples, fully synthetic and disjoint from every evaluation set). Sizes above are listed `coco / monet`.
+
+Because the auxiliary store and the FAISS index are shared between the base and large variants, **one bank is four files**, and the only difference between CorrCLIPv2-B and -L is `metaclip_b` vs `metaclip_l`.
+
+## Sample Usage
+
+Masks — extract into `data/`, which yields `data/instance_mask/{voc,context,coco,ade,ade847,city}/`:
+
+```bash
+wget https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/instance_mask.zip
+unzip instance_mask.zip -d data/
+```
+
+Memory bank — keep the directory layout, the loader resolves each file by name:
+
+```bash
+mkdir -p memory_bank/clip_embeddings memory_bank/vfm_embeddings
+BASE=https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank
+wget -c -P memory_bank/clip_embeddings $BASE/clip_embeddings/coco_metaclip_l.pt
+wget -c -P memory_bank/clip_embeddings $BASE/clip_embeddings/coco_dfnclip_b.pt
+wget -c -P memory_bank/vfm_embeddings  $BASE/vfm_embeddings/coco_radiov3.faiss
+```
+
+Then set `_clip_scale = 'large'` (or `'base'`) and `_mb_bank = 'coco'` (or `'monet'`) in `configs/base_config.py` and evaluate as usual. Keep `vfm_type = 'radiov3'` — that is the feature space these indexes were built in, and the queries must share it.
+
+## Source data
+
+The masks are derived from the public validation images of the corresponding benchmarks; the `coco` bank is derived from COCO train2017 and the `monet` bank from the MONET Z-Image synthetic subset. Only derived features are redistributed here, never the source images — the original datasets' own licenses and terms continue to apply to them.
+
+## Citation
+
+The CorrCLIPv2 journal extension is under review; please cite the conference paper for now:
+
+```bibtex
+@article{zhang2024corrclip,
+  title={Corrclip: Reconstructing patch correlations in clip for open-vocabulary semantic segmentation},
+  author={Zhang, Dengke and Liu, Fagui and Tang, Quan},
+  journal={arXiv preprint arXiv:2411.10086},
+  year={2024}
+}
+```