Mirrors/ECCV2022-RIFE
1
0
Fork 0
mirror of https://github.com/hzwer/ECCV2022-RIFE.git synced 2026-02-24 04:19:41 +01:00
Code Issues Packages Projects Releases Wiki Activity

Update distill loss

Browse Source
This commit is contained in:
hzwer
2022-03-05 12:34:26 +08:00
parent fbfbdf0459
commit 7788494700
3 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
model/IFNet.py
View File

@@ -99,7 +99,7 @@ class IFNet(nn.Module):
merged[i] = merged[i][0] * mask_list[i] + merged[i][1] * (1 - mask_list[i])
if gt.shape[1] == 3:
loss_mask = ((merged[i] - gt).abs().mean(1, True) > (merged_teacher - gt).abs().mean(1, True) + 0.01).float().detach()
loss_distill += ((flow_teacher.detach() - flow_list[i]).abs() * loss_mask).mean()
loss_distill += (((flow_teacher.detach() - flow_list[i]) ** 2).mean(1, True) ** 0.5) * loss_mask).mean()
c0 = self.contextnet(img0, flow[:, :2])
c1 = self.contextnet(img1, flow[:, 2:4])
tmp = self.unet(img0, img1, warped_img0, warped_img1, mask, flow, c0, c1)

2
model/IFNet_m.py
View File

@@ -100,7 +100,7 @@ class IFNet_m(nn.Module):
merged[i] = merged[i][0] * mask_list[i] + merged[i][1] * (1 - mask_list[i])
if gt.shape[1] == 3:
loss_mask = ((merged[i] - gt).abs().mean(1, True) > (merged_teacher - gt).abs().mean(1, True) + 0.01).float().detach()
loss_distill += ((flow_teacher.detach() - flow_list[i]).abs() * loss_mask).mean()
loss_distill += (((flow_teacher.detach() - flow_list[i]) ** 2).mean(1, True) ** 0.5) * loss_mask).mean()
c0 = self.contextnet(img0, flow[:, :2])
c1 = self.contextnet(img1, flow[:, 2:4])
tmp = self.unet(img0, img1, warped_img0, warped_img1, mask, flow, c0, c1)

4
train.py
View File

@@ -21,10 +21,10 @@ log_path = 'train_log'
def get_learning_rate(step):
if step < 2000:
mul = step / 2000.
return 3e-4 * mul
return 1e-4 * mul
else:
mul = np.cos((step - 2000) / (args.epoch * args.step_per_epoch - 2000.) * math.pi) * 0.5 + 0.5
return (3e-4 - 3e-5) * mul + 3e-5
return (1e-4 - 1e-5) * mul + 1e-5
def flow2rgb(flow_map_np):
h, w, _ = flow_map_np.shape