search

TEA-Lab / diffusion_reward

[ECCV 2024] 💐Official implementation of the paper "Diffusion Reward: Learning Rewards via Conditional Video Diffusion"
https://diffusion-reward.github.io/
MIT License
70 stars 7 forks source link

About Quantitative comparison of video models #8

Closed yingchengyang closed 2 months ago

yingchengyang commented 3 months ago

Thanks a lot for such a wonderful work! I'm curious about the Quantitative comparison of different video models (like videogpt and vqdiffusion) in table 1 of the paper. How can I evaluate the performance of video models? Thanks a lot!

luccachiang commented 3 months ago

Hi! Thanks for your interest. The results are calculated by generated videos and their ground truths. In our case, we give one video, i.e. ground truth, as a condition to the model to generate one corresponding video. As for metric calculation, please kindly refer to this repo for more details :)