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