This repository provides a randomly selected sample from the full benchmark, using the same folder and file conventions as the complete release.
images/<sequence_id>/
RGB frames (00000.jpg = query frame; 00001.jpg … = search video).
masks/<sequence_id>/
PNG masks aligned by frame stem (sparse where applicable).
train_t-IoU.txt
Training split IDs in this sample (9 sequences).… See the full description on the dataset page:
https://huggingface.co/datasets/VqsReseacher/VQS-4k-sample.