From 6ffc1c06f66277d37877fc13fb1ffa585598d6d7 Mon Sep 17 00:00:00 2001
From: Jordan Gong <jordan.gong@protonmail.com>
Date: Thu, 14 Jan 2021 23:43:29 +0800
Subject: Enable optimizer fine tuning
---
models/model.py | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)
(limited to 'models')
diff --git a/models/model.py b/models/model.py
index 740cdf3..8797636 100644
--- a/models/model.py
+++ b/models/model.py
@@ -133,12 +133,21 @@ class Model:
dataloader = self._parse_dataloader_config(dataset, dataloader_config)
# Prepare for model, optimizer and scheduler
model_hp = self.hp.get('model', {})
- optim_hp = self.hp.get('optimizer', {})
+ optim_hp: dict = self.hp.get('optimizer', {}).copy()
+ ae_optim_hp = optim_hp.pop('auto_encoder', {})
+ pn_optim_hp = optim_hp.pop('part_net', {})
+ hpm_optim_hp = optim_hp.pop('hpm', {})
+ fc_optim_hp = optim_hp.pop('fc', {})
sched_hp = self.hp.get('scheduler', {})
self.rgb_pn = RGBPartNet(self.train_size, self.in_channels, **model_hp)
# Try to accelerate computation using CUDA or others
self.rgb_pn = self.rgb_pn.to(self.device)
- self.optimizer = optim.Adam(self.rgb_pn.parameters(), **optim_hp)
+ self.optimizer = optim.Adam([
+ {'params': self.rgb_pn.ae.parameters(), **ae_optim_hp},
+ {'params': self.rgb_pn.pn.parameters(), **pn_optim_hp},
+ {'params': self.rgb_pn.hpm.parameters(), **hpm_optim_hp},
+ {'params': self.rgb_pn.fc_mat, **fc_optim_hp},
+ ], **optim_hp)
self.scheduler = optim.lr_scheduler.StepLR(self.optimizer, **sched_hp)
self.writer = SummaryWriter(self._log_name)
--
cgit v1.2.3
From 59ccfd7718babe94fac549fcfbfa22bb311f0bd8 Mon Sep 17 00:00:00 2001
From: Jordan Gong <jordan.gong@protonmail.com>
Date: Thu, 21 Jan 2021 21:50:28 +0800
Subject: Bug fixes
1. Turn off autograd while decoding canonical and pose features
2. Change default batch size to (4, 8)
---
models/auto_encoder.py | 19 ++++++++++---------
1 file changed, 10 insertions(+), 9 deletions(-)
(limited to 'models')
diff --git a/models/auto_encoder.py b/models/auto_encoder.py
index 5e7558b..36be868 100644
--- a/models/auto_encoder.py
+++ b/models/auto_encoder.py
@@ -132,15 +132,16 @@ class AutoEncoder(nn.Module):
# x_c1_t2 is the frame for later module
(f_a_c1_t2, f_c_c1_t2, f_p_c1_t2) = self.encoder(x_c1_t2)
- # Decode canonical features for HPM
- x_c_c1_t2 = self.decoder(
- torch.zeros_like(f_a_c1_t2), f_c_c1_t2, torch.zeros_like(f_p_c1_t2),
- no_trans_conv=True
- )
- # Decode pose features for Part Net
- x_p_c1_t2 = self.decoder(
- torch.zeros_like(f_a_c1_t2), torch.zeros_like(f_c_c1_t2), f_p_c1_t2
- )
+ with torch.no_grad():
+ # Decode canonical features for HPM
+ x_c_c1_t2 = self.decoder(
+ torch.zeros_like(f_a_c1_t2), f_c_c1_t2, torch.zeros_like(f_p_c1_t2),
+ no_trans_conv=True
+ )
+ # Decode pose features for Part Net
+ x_p_c1_t2 = self.decoder(
+ torch.zeros_like(f_a_c1_t2), torch.zeros_like(f_c_c1_t2), f_p_c1_t2
+ )
if self.training:
# t1 is random time step, c2 is another condition
--
cgit v1.2.3
From 04c9d3210ff659bbe00dedb2d193a748e7a97b54 Mon Sep 17 00:00:00 2001
From: Jordan Gong <jordan.gong@protonmail.com>
Date: Thu, 21 Jan 2021 23:32:53 +0800
Subject: Print average losses after 100 iters
---
models/model.py | 20 +++++++++++++-------
models/rgb_part_net.py | 5 ++---
2 files changed, 15 insertions(+), 10 deletions(-)
(limited to 'models')
diff --git a/models/model.py b/models/model.py
index 8797636..3b54363 100644
--- a/models/model.py
+++ b/models/model.py
@@ -164,6 +164,9 @@ class Model:
# Training start
start_time = datetime.now()
+ running_loss = torch.zeros(4).to(self.device)
+ print(f"{'Iter':^5} {'Loss':^6} {'Xrecon':^8} {'PoseSim':^8}",
+ f"{'CanoCons':^8} {'BATrip':^8} {'LR':^9}")
for (batch_c1, batch_c2) in dataloader:
self.curr_iter += 1
# Zero the parameter gradients
@@ -172,24 +175,27 @@ class Model:
x_c1 = batch_c1['clip'].to(self.device)
x_c2 = batch_c2['clip'].to(self.device)
y = batch_c1['label'].to(self.device)
- loss, metrics = self.rgb_pn(x_c1, x_c2, y)
+ losses = self.rgb_pn(x_c1, x_c2, y)
+ loss = losses.sum()
loss.backward()
self.optimizer.step()
# Step scheduler
self.scheduler.step()
+ # Statistics and checkpoint
+ running_loss += losses.detach()
# Write losses to TensorBoard
- self.writer.add_scalar('Loss/all', loss.item(), self.curr_iter)
+ self.writer.add_scalar('Loss/all', loss, self.curr_iter)
self.writer.add_scalars('Loss/details', dict(zip([
'Cross reconstruction loss', 'Pose similarity loss',
'Canonical consistency loss', 'Batch All triplet loss'
- ], metrics)), self.curr_iter)
+ ], losses)), self.curr_iter)
if self.curr_iter % 100 == 0:
- print('{0:5d} loss: {1:6.3f}'.format(self.curr_iter, loss),
- '(xrecon = {:f}, pose_sim = {:f},'
- ' cano_cons = {:f}, ba_trip = {:f})'.format(*metrics),
- 'lr:', self.scheduler.get_last_lr()[0])
+ print(f'{self.curr_iter:5d} {running_loss.sum() / 100:6.3f}',
+ '{:f} {:f} {:f} {:f}'.format(*running_loss / 100),
+ f'{self.scheduler.get_last_lr()[0]:.3e}')
+ running_loss.zero_()
if self.curr_iter % 1000 == 0:
torch.save({
diff --git a/models/rgb_part_net.py b/models/rgb_part_net.py
index f39b40b..e707c26 100644
--- a/models/rgb_part_net.py
+++ b/models/rgb_part_net.py
@@ -80,9 +80,8 @@ class RGBPartNet(nn.Module):
if self.training:
batch_all_triplet_loss = self.ba_triplet_loss(x, y)
- losses = (*losses, batch_all_triplet_loss)
- loss = torch.sum(torch.stack(losses))
- return loss, [loss.item() for loss in losses]
+ losses = torch.stack((*losses, batch_all_triplet_loss))
+ return losses
else:
return x.unsqueeze(1).view(-1)
--
cgit v1.2.3
From d750dd9dafe3cda3b1331ad2bfecb53c8c2b1267 Mon Sep 17 00:00:00 2001
From: Jordan Gong <jordan.gong@protonmail.com>
Date: Thu, 21 Jan 2021 23:47:11 +0800
Subject: A type hint fix
---
models/model.py | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
(limited to 'models')
diff --git a/models/model.py b/models/model.py
index bed28a5..f4604c8 100644
--- a/models/model.py
+++ b/models/model.py
@@ -133,7 +133,7 @@ class Model:
dataloader = self._parse_dataloader_config(dataset, dataloader_config)
# Prepare for model, optimizer and scheduler
model_hp = self.hp.get('model', {})
- optim_hp: dict = self.hp.get('optimizer', {}).copy()
+ optim_hp: Dict = self.hp.get('optimizer', {}).copy()
ae_optim_hp = optim_hp.pop('auto_encoder', {})
pn_optim_hp = optim_hp.pop('part_net', {})
hpm_optim_hp = optim_hp.pop('hpm', {})
--
cgit v1.2.3