Molmo2-VideoCountEval is a validation set of video pointing data collected from human annotators.
It can be used to evaluate vision-language models' video counting capability.
Molmo2-VideoCountEval is part of the Molmo2 dataset collection and was used to evaluate the Molmo2 family of models.
Quick links:
video_source: There are two video sources: youtube, and MammalNet. For YouTube videos, you need to… See the full description on the dataset page:
https://huggingface.co/datasets/allenai/Molmo2-VideoCountEval.
