Commit @7cbe1b2dee697674c26dd8b7f95db7c44c354532

윤영준 2023-07-05

Generator visualizer

@7cbe1b2dee697674c26dd8b7f95db7c44c354532

3a11ad1

7cbe1b2

train.py

--- train.py

+++ train.py


 ARNN_loss_window = vis.line(Y=np.array([0]), X=np.array([0]), opts=dict(title='Generator-AttentionRNN Loss'))
 AE_loss_window = vis.line(Y=np.array([0]), X=np.array([0]), opts=dict(title='Generator-AutoEncoder Loss'))
 Discriminator_loss_window = vis.line(Y=np.array([0]), X=np.array([0]), opts=dict(title='Discriminator Loss'))
-attenton_map_visualizer = vis.image(np.zeros((692,776)), opts=dict(title='Attention Map'))
-
+Attention_map_visualizer = vis.image(np.zeros((692, 776)), opts=dict(title='Attention Map'))
+Generator_output_visualizer = vis.image(np.zeros((692,776)), opts=dict(title='Generated Derain Output'))
 
 for epoch_num, epoch in enumerate(range(epochs)):
     for i, imgs in enumerate(dataloader):

         }
 
         logger.print_training_log(epoch_num, epochs, i, len(dataloader), losses)
-
         # visdom logger
         vis.line(Y=np.array([generator_loss_ARNN.item()]), X=np.array([epoch * epoch_num + i]), win=ARNN_loss_window,
                  update='append')

                  update='append')
         vis.line(Y=np.array([discriminator_loss.item()]), X=np.array([epoch * epoch_num + i]), win=Discriminator_loss_window,
                  update='append')
-        vis.image(generator_attention_map[-1], win=attenton_map_visualizer, opts=dict(title="Attention Map"))
-
+        vis.image(generator_attention_map[-1][0,0,:,:], win=Attention_map_visualizer, opts=dict(title="Attention Map"))
+        vis.image(generator_result['skip_3'][-1][0,0,:,:], win=Generator_output_visualizer, opts=dict(title="Generator Output"))
     day = strftime("%Y-%m-%d %H:%M:%S", gmtime())
     if epoch % save_interval == 0 and epoch != 0:
         torch.save(generator.attentiveRNN.state_dict(), f"weight/Attention_RNN_{epoch}_{day}.pt")

Add a comment

Open 0
Closed 0

List

...	...	@@ -91,8 +91,8 @@
91	91	ARNN_loss_window = vis.line(Y=np.array([0]), X=np.array([0]), opts=dict(title='Generator-AttentionRNN Loss'))
92	92	AE_loss_window = vis.line(Y=np.array([0]), X=np.array([0]), opts=dict(title='Generator-AutoEncoder Loss'))
93	93	Discriminator_loss_window = vis.line(Y=np.array([0]), X=np.array([0]), opts=dict(title='Discriminator Loss'))
94		-attenton_map_visualizer = vis.image(np.zeros((692,776)), opts=dict(title='Attention Map'))
95		-
	94	+Attention_map_visualizer = vis.image(np.zeros((692, 776)), opts=dict(title='Attention Map'))
	95	+Generator_output_visualizer = vis.image(np.zeros((692,776)), opts=dict(title='Generated Derain Output'))
96	96
97	97	for epoch_num, epoch in enumerate(range(epochs)):
98	98	for i, imgs in enumerate(dataloader):
...	...	@@ -146,7 +146,6 @@
146	146	}
147	147
148	148	logger.print_training_log(epoch_num, epochs, i, len(dataloader), losses)
149		-
150	149	# visdom logger
151	150	vis.line(Y=np.array([generator_loss_ARNN.item()]), X=np.array([epoch * epoch_num + i]), win=ARNN_loss_window,
152	151	update='append')
...	...	@@ -154,8 +153,8 @@
154	153	update='append')
155	154	vis.line(Y=np.array([discriminator_loss.item()]), X=np.array([epoch * epoch_num + i]), win=Discriminator_loss_window,
156	155	update='append')
157		- vis.image(generator_attention_map[-1], win=attenton_map_visualizer, opts=dict(title="Attention Map"))
158		-
	156	+ vis.image(generator_attention_map[-1][0,0,:,:], win=Attention_map_visualizer, opts=dict(title="Attention Map"))
	157	+ vis.image(generator_result['skip_3'][-1][0,0,:,:], win=Generator_output_visualizer, opts=dict(title="Generator Output"))
159	158	day = strftime("%Y-%m-%d %H:%M:%S", gmtime())
160	159	if epoch % save_interval == 0 and epoch != 0:
161	160	torch.save(generator.attentiveRNN.state_dict(), f"weight/Attention_RNN_{epoch}_{day}.pt")

Delete comment