Skip to content

Fix bug in stage1 (en-/decoder pre-) training #15

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
sRassmann wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix bug in stage1 (en-/decoder pre-) training #15

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 20 additions & 13 deletions VQ-GAN/taming/models/vqgan.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# fmt: off
import random
import torch
import torch.nn as nn
Expand Down Expand Up @@ -76,25 +77,28 @@ def decode_code(self, code_b):
dec = self.decode(quant_b)
return dec

def forward(self, input, target):
def forward(self, input, target=None):
quant, diff, _ = self.encode(input)
if target is not None: quant = self.spade(quant, target)
dec = self.decode(quant)
return dec, diff

def get_input(self, batch, k):
x = batch[k]

return x.float()

def training_step(self, batch, batch_idx, optimizer_idx):
source = random.choice(self.modalities)
target = random.choice(self.modalities)
if self.stage == 1:
target = source
else:
target = random.choice(self.modalities)
x_src = self.get_input(batch, source)
x_tar = self.get_input(batch, target)
skip_pass = 0

if self.stage == 1:
if self.stage == 1:
xrec, qloss = self(x_src)
else:
z_src, qloss, _ = self.encode(x_src)
Expand Down Expand Up @@ -122,11 +126,14 @@ def training_step(self, batch, batch_idx, optimizer_idx):

def validation_step(self, batch, batch_idx):
source = random.choice(self.modalities)
target = random.choice(self.modalities)
if self.stage == 1:
target = source
else:
target = random.choice(self.modalities)
x_src = self.get_input(batch, source)
x_tar = self.get_input(batch, target)
if self.stage == 1:

if self.stage == 1:
xrec, qloss = self(x_src)
else:
z_src, qloss, _ = self.encode(x_src)
Expand All @@ -141,10 +148,10 @@ def validation_step(self, batch, batch_idx):
discloss, log_dict_disc = self.loss(qloss, x_tar, xrec, 1, self.global_step,
last_layer=self.get_last_layer(), split="val")
rec_loss = log_dict_ae["val/rec_loss"]
self.log("val/rec_loss", rec_loss,
prog_bar=True, logger=True, on_step=True, on_epoch=True, sync_dist=True)
self.log("val/aeloss", aeloss,
prog_bar=True, logger=True, on_step=True, on_epoch=True, sync_dist=True)
# self.log("val/rec_loss", rec_loss,
# prog_bar=True, logger=True, on_step=True, on_epoch=True, sync_dist=True)
# self.log("val/aeloss", aeloss,
# prog_bar=True, logger=True, on_step=True, on_epoch=True, sync_dist=True)
self.log_dict(log_dict_ae)
self.log_dict(log_dict_disc)
return self.log_dict
Expand Down Expand Up @@ -188,7 +195,7 @@ def log_images(self, batch, **kwargs):
xrec = self.to_rgb(xrec)
log["source"] = x_src
log["target"] = x_tar
if self.stage == 1:
if self.stage == 1:
log["recon"] = xrec
else:
log[f"recon_{source}_to_{target}"] = xrec
Expand Down Expand Up @@ -448,4 +455,4 @@ def configure_optimizers(self):
lr=lr, betas=(0.5, 0.9))
opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
lr=lr, betas=(0.5, 0.9))
return [opt_ae, opt_disc], []
return [opt_ae, opt_disc], []