README.md

---
task_categories:
  - image-segmentation
license: cc-by-nc-4.0
library_name: mmsegmentation
---

# CorrCLIPv2: Pre-generated Masks and Prebuilt Memory Banks

This repository provides the pre-generated region masks and prebuilt memory banks for **CorrCLIPv2: Harnessing Unlabeled Images for Open-Vocabulary Semantic Segmentation**, published in the **International Journal of Computer Vision (IJCV), 2026**.

[Journal version](https://link.springer.com/article/10.1007/s11263-026-03027-3) · [arXiv version](https://arxiv.org/abs/2610.24757)

The arXiv version corresponds to the initial submission, while the journal version includes additional ablation studies and re-evaluates other methods using the same models as CorrCLIPv2 for a consistent comparison.

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 [CorrCLIPv2 README](https://github.com/zdk258/CorrCLIP/blob/master/CorrCLIPv2/README.md). Run commands from the `CorrCLIPv2/` directory unless otherwise specified; all local paths below are relative to that directory.

## Region masks

[instance_mask.zip](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/instance_mask.zip?download=true) (110 MiB) — EntitySeg region masks for the ten standard benchmarks. Extract it into `data/`, which gives `data/instance_mask/{voc,context,coco,ade,ade847,city}/` as `instance_mask_path` expects. With these in place, set `mask_generator=None` and no mask generator needs to be installed.

## Prebuilt banks

Four banks are provided: `coco`, built from COCO train2017 and used for the main results; `monet`, built from the fully synthetic MONET Z-Image subset; and `gpic` and `sa1b`, built from GPIC and SA-1B images. The latter three banks are image-level disjoint from the evaluation sets. Prebuilt files support both CorrCLIPv2-B and CorrCLIPv2-L for `coco` and `monet`, and CorrCLIPv2-B for `gpic` and `sa1b`.

With DMCE enabled, one configuration needs three files: a main store matching the CLIP scale, the DFN auxiliary store, and the FAISS index. Pick a row and download those files; for a given bank, the B and L configurations share the auxiliary store and index. Sizes below are approximate binary units (GiB/MiB).

| Configuration | Files to download | Total |
|---|---|---|
| **CorrCLIPv2-L** + `coco` (main results) | [coco_metaclip_l.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/coco_metaclip_l.pt?download=true) 4.22 GiB · [coco_dfnclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/coco_dfnclip_b.pt?download=true) 2.81 GiB · [coco_radiov3.faiss](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/vfm_embeddings/coco_radiov3.faiss?download=true) 190 MiB | 7.21 GiB |
| **CorrCLIPv2-B** + `coco` | [coco_metaclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/coco_metaclip_b.pt?download=true) 2.81 GiB · [coco_dfnclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/coco_dfnclip_b.pt?download=true) 2.81 GiB · [coco_radiov3.faiss](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/vfm_embeddings/coco_radiov3.faiss?download=true) 190 MiB | 5.81 GiB |
| **CorrCLIPv2-L** + `monet` (synthetic, disjoint) | [monet_metaclip_l.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/monet_metaclip_l.pt?download=true) 2.15 GiB · [monet_dfnclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/monet_dfnclip_b.pt?download=true) 1.43 GiB · [monet_radiov3.faiss](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/vfm_embeddings/monet_radiov3.faiss?download=true) 108 MiB | 3.68 GiB |
| **CorrCLIPv2-B** + `monet` | [monet_metaclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/monet_metaclip_b.pt?download=true) 1.43 GiB · [monet_dfnclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/monet_dfnclip_b.pt?download=true) 1.43 GiB · [monet_radiov3.faiss](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/vfm_embeddings/monet_radiov3.faiss?download=true) 108 MiB | 2.97 GiB |
| **CorrCLIPv2-B** + `gpic` (disjoint) | [gpic_metaclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/gpic_metaclip_b.pt?download=true) 2.20 GiB · [gpic_dfnclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/gpic_dfnclip_b.pt?download=true) 2.20 GiB · [gpic_radiov3.faiss](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/vfm_embeddings/gpic_radiov3.faiss?download=true) 154 MiB | 4.54 GiB |
| **CorrCLIPv2-B** + `sa1b` (disjoint) | [sa1b_metaclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/sa1b_metaclip_b.pt?download=true) 3.69 GiB · [sa1b_dfnclip_b.pt](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/clip_embeddings/sa1b_dfnclip_b.pt?download=true) 3.69 GiB · [sa1b_radiov3.faiss](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/vfm_embeddings/sa1b_radiov3.faiss?download=true) 240 MiB | 7.61 GiB |

Put the `.pt` files in `memory_bank/clip_embeddings/` and the `.faiss` file in `memory_bank/vfm_embeddings/`, keeping the filenames unchanged — the loader resolves each by name. Then set `_clip_scale` and `_mb_bank` in `configs/base_config.py` to match the row you chose and run as usual; `vfm_type` stays `radiov3`, the space these indexes were built in. For the no-DMCE ablation, set `use_aux_model=False` and skip the DFN store.

Each `.pt` store is an `[N, D]` fp16 tensor loaded fully into RAM at startup; its row index is the region id, and the FAISS index (IDMap2) returns the same ids, so search results index the stores directly.

## Bank image lists

The image lists behind the disjoint banks are released for auditing — one source image path per line, the exact set each bank was built from: [gpic](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/bank_images_gpic.txt?download=true) (6.2 MiB), [monet](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/bank_images_monet.txt?download=true) (2.8 MiB), [sa1b](https://huggingface.co/datasets/dk258/CorrCLIPv2/resolve/main/memory_bank/bank_images_sa1b.txt?download=true) (1.0 MiB). The `coco` bank needs no list: it is the full COCO train2017 image set.

## 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. This repository distributes derived masks, feature stores, retrieval indexes, and source image lists, but no source images — the original datasets' own licenses and terms continue to apply to them. The disjoint-bank source images are available from [SA-1B](https://ai.meta.com/datasets/segment-anything/), [GPIC](https://huggingface.co/datasets/stanford-vision-lab/gpic), and [MONET](https://huggingface.co/datasets/jasperai/monet).

## Citation

If you use CorrCLIPv2 or these artifacts in your research, please cite the journal paper:

```bibtex
@article{zhang2026corrclipv2,
    author    = {Zhang, Dengke and Liu, Fagui and Tang, Quan and Chen, C. L. Philip},
    title     = {{CorrCLIPv2}: Harnessing Unlabeled Images for Open-Vocabulary Semantic Segmentation},
    journal   = {International Journal of Computer Vision},
    year      = {2026},
    volume    = {134},
    number    = {11},
    pages     = {478},
    publisher = {Springer},
    doi       = {10.1007/s11263-026-03027-3}
}
```