search

andrewjong / ShineOn-Virtual-Tryon

Code for "ShineOn: Illuminating Design Choices for Practical Video-based Virtual Clothing Try-on", accepted at WACV 2021 Generation of Human Behavior Workshop.
131 stars 31 forks source link

Flownet2 with PT1.6! Resampling with SAMS, UNet. Fixed Data Leakage! #87

Closed gauravkuppa closed 4 years ago

gauravkuppa commented 4 years ago

FlowNet2 works with PT1.6 Resample with SAMS, #67 Fixed data leakage issue with UNet Multi-frame training works for both SAMS and UNet Acitvation function manipulation from commandline option Changes to UNet visualization, only visualizes latest frame image

Note: you need to use at least 2 frames for UNet and SAMS to use flow; otherwise, you are resampling with a zero tensor

gauravkuppa commented 4 years ago

Tested successfully using these two commands

python train.py --name DELETEME_pt1.6_sams --model sams --batch 2 --ngf_power_start 3 --ngf_power_end 3 --n_frames_total 5 --n_frames_now 5  -j 8 --gan_mode hinge --datacap 200 --datacap_val 20 --display_count 50 --vvt_dataroot ~/data/fw_gan_vvt --gpu_ids 1 --flow_warp --activation swish
python train.py --name DELETEME_pt1.6_sams --model unet --batch 2 --n_frames_total 5 --n_frames_now 5  -j 8 --datacap 200 --datacap_val 20 --display_count 50 --vvt_dataroot ~/data/fw_gan_vvt --gpu_ids 0,1 --flow_warp