search

amjltc295 / Free-Form-Video-Inpainting

Official Pytorch implementation of "Learnable Gated Temporal Shift Module for Deep Video Inpainting. Chang et al. BMVC 2019." and the FVI dataset in "Free-form Video Inpainting with 3D Gated Convolution and Temporal PatchGAN, Chang et al. ICCV 2019"
https://arxiv.org/abs/1907.01131
335 stars 53 forks source link

Can't reproduce inference results for LGTSM in paper #24

Closed Mykheievskyi closed 4 years ago

Mykheievskyi commented 4 years ago

Thanks for sharing your code. I try to reproduce inference results with pre-trained weights (weights for the BMVC 2019 work) and get only 7 FPS on GTX 1080Ti. In the paper you report 80 FPS for LGTMS solution, can you tell me what conditions these results were obtained?

amjltc295 commented 4 years ago

For all models, we use a single Tesla V100 GPU with a Intel(R) Xeon(R) CPU E5-2650. We run the FVI dataset where video length is 15 frames at most. I forgot the batch size (should be 1~3), but you could test the relative time compared to other models and it should be accurate as we compare them with the same setting.