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

Training Time for ICCV 2019 Method #18

Closed harsh-99 closed 4 years ago

harsh-99 commented 4 years ago

Hi,

Thanks for sharing your code. I am working on the same dataset FVI. I just wanted to know for how long did you trained the network and you have mentioned that you have used 1940 video and then applied a few data augmentation techniques. Did you apply data augmentation on all the videos? Can you please mention the final size of the dataset after augmentation? Also, It would be a great help if you can mention the weight of different losses (i.e. perceptual, style, reconstruction and adversarial) you used while training.

Looking for your help

amjltc295 commented 4 years ago
  1. The free-form mask itself is the augmentation. There are only 1940 videos for training. For other augmentation, please refer to the code.
  2. About 1/1/10/120/0.1 for recon/masked-recon/perc/style/adv.