This repository redistributes a PREPROCESSED version (HDF5 packed body crops and face crops) of the third-party CCVID person re-identification dataset, which is released by its original authors under a research-only license/agreement. By requesting access you confirm that (1) you have already obtained the original CCVID dataset from its official source and agreed to its terms, (2) you will use these files for non-commercial academic research only, (3) you will not redistribute them, and (4) you will cite the original dataset and our paper in any publication.

CCVID H5: preprocessed CCVID with body and face crops

Preprocessed HDF5 pack of the CCVID clothes-changing video person re-identification benchmark, as used in our paper FusionAgent: A Multimodal Agent with Dynamic Model Selection for Human Recognition (arXiv:2603.26908, code).

File	Size	Content
`ccvid.h5`	25.2 GB	body frames, `(256,128,3)`
`ccvid_face.h5`	11.8 GB	face crops, `(112,112,3)`
`train.txt`, `query.txt`, `gallery.txt`	<1 MB	official splits, unchanged (`video_path pid clothes_label`)
`readme.txt`		original CCVID readme (license + citation)

Download

pip install -U huggingface_hub
hf auth login   # you must have been granted access to this gated repo
hf download Paipile/ccvid-h5 --repo-type dataset --local-dir ./data/CCVID

Point a CAL-style CCVID loader at ./data/CCVID; the split files are read exactly as in the original dataset and the video path in each entry is the H5 group key.

H5 format

Keys mirror the directory structure of the original dataset; <img> is the original image file name without extension. Every leaf is an HDF5 dataset holding one RGB uint8 array of shape (H, W, 3) (i.e. np.asarray(PIL.Image.open(...).convert('RGB'))). Face crops are aligned 112×112 crops detected inside each body image; face_<k> enumerates multiple detections in the same image (most images have a single face_0). Images without a detectable face are simply absent from the face file.

Open the files with h5py.File(path, 'r') and keep the handle open across __getitem__ calls (one handle per worker when using multiprocess DataLoader workers). The arrays are stored uncompressed for fast random access.

File	Key path	Shape
`ccvid.h5`	`session{1,2,3}/<pid>_<clip>/<frame>`	`(256,128,3)`
`ccvid_face.h5`	`session{1,2,3}/<pid>_<clip>/<frame>/face_<k>`	`(112,112,3)`

import h5py
from PIL import Image
body = h5py.File('data/CCVID/ccvid.h5', 'r'); face = h5py.File('data/CCVID/ccvid_face.h5', 'r')
tracklet = 'session1/001_01'
frames = sorted(body[tracklet].keys())                     # same order as listing the folder
img = Image.fromarray(body[f'{tracklet}/{frames[0]}'][:])   # body frame
key = f'{tracklet}/{frames[0]}'
if key in face:                                            # not every frame has a face
    faces = [Image.fromarray(face[key][k][:]) for k in face[key]]

License and terms

This repository contains derived data only; all rights to the images remain with the original authors. CCVID (Gu et al., Clothes-Changing Person Re-identification with RGB Modality Only, CVPR 2022) is built from FVG (Zhang et al., CVPR 2019) and released under CC BY-NC-SA 4.0; see readme.txt and https://github.com/guxinqian/Simple-CCReID. You must obtain the original dataset and accept its terms before using these files. Access here is gated: non-commercial academic research only, no redistribution.

Citation

If you use these files, please cite the original dataset above and our paper:

@article{zhu2026fusionagent,
  title   = {FusionAgent: A Multimodal Agent with Dynamic Model Selection for Human Recognition},
  author  = {Zhu, Jie and Guo, Xiao and Su, Yiyang and Jain, Anil and Liu, Xiaoming},
  journal = {arXiv preprint arXiv:2603.26908},
  year    = {2026}
}

Downloads last month: 24

Total file size:

37 GB

Paper for Paipile/ccvid-h5

FusionAgent: A Multimodal Agent with Dynamic Model Selection for Human Recognition

Paper • 2603.26908 • Published Mar 27