For videos eval

[Yingshu-Li]

momentseeker_1k8:
dataset_parser: momentseeker
dataset_name: momentseeker_1k8
video_root: ""
clip_root: ""
frame_root: /MMEB/data/vlm2vec_eval/video-tasks/frames/video_mret/MomentSeeker
num_negative_clips: 9
num_video_frames: 8
eval_type: local

Do we still need this one, as when I include this one, I will get this error: KeyError: 'momentseeker_1k8'

[XMHZZ2018]

@Yingshu-Li

Thanks for the question. The momentseeker_1k8 version is a backup and archived copy. For context, the original MomentSeeker dataset (which contained 1.8k samples) had some duplicate entries. After the original authors later fixed these issues, we adopted the deduplicated version (which contains 1.6k samples).

We keep the separate momentseeker_1k8 tab in case some users still rely on the older version.

For more details, please refer to our changelog: https://github.com/TIGER-AI-Lab/VLM2Vec/blob/main/CHANGELOG.md.