save_out = output
netD.zero_grad()
errD_real = criterion(output, labelv)
errD_real.backward()
D_x = errD_real.data.mean()
optimizerD.step()
# train with fake
if cv2.waitKey(12) & 0xFF == ord('q'):
break
print('[%d/%d][%d/%d] Loss_D: %.4f Loss_G: %.4f D(x): %.4f D(G(z)): %.4f / %.4f'
% (epoch, 0, read_id, 0,
errD_real.data, D_x, 0, 0, 0))
if Visdom_flag == True:
plotter.plot( 'LOSS', 'LOSS', 'LOSS', iteration_num, D_x.cpu().detach().numpy())
if read_id % 2 == 0:
#vutils.save_image(real_cpu,
# '%s/real_samples.png' % opt.outf,
# normalize=True)
#netG.eval()
#fake = netG(fixed_noise)
#cv2.imwrite(Save_pic_dir + str(i) +".jpg", mat)
#show the result
dispay_id =0
Matrix = mydata_loader.input_mat[dispay_id,0,:,:] +104
show1 = Matrix*0
path2 = (mydata_loader.input_path[dispay_id,:])*Original_window_Len
show1[int(path2[0]),:]=254
#optimizerG.step()
#save_out = Fake
# train with fake
# if cv2.waitKey(12) & 0xFF == ord('q'):
# break
print (" all test point time is [%f] " % ( test_time_point - start_time))
if validation_flag==False:
print('[%d/%d][%d/%d] Loss_D: %.4f Loss_G: %.4f D(x): %.4f D(G(z)): %.4f / %.4f'
% (epoch, 0, read_id, 0,
G_x, D_x, 0, 0, 0))
if read_id % 2 == 0 and Visdom_flag == True and validation_flag==False:
plotter.plot( 'DLOSS', 'DLOSS', 'DLOSS', iteration_num, D_x.cpu().detach().numpy())
plotter.plot( 'GLOSS', 'GLOSS', 'GLOSS', iteration_num, G_x.cpu().detach().numpy())
plotter.plot( 'GlLOSS', 'GlLOSS', 'GlLOSS', iteration_num, G_x_L12.cpu().detach().numpy())
#plotter.plot( 'cLOSS1', 'cLOSS1', 'cLOSS1', iteration_num, D_x1.cpu().detach().numpy())
#plotter.plot( 'cLOSS12', 'cLOSS2', 'cLOSS2', iteration_num, D_x2.cpu().detach().numpy())
#plotter.plot( 'cLOSS_f', 'cLOSSf', 'cLOSSf', iteration_num, D_xf.cpu().detach().numpy())
if read_id % 1 == 0 and Display_fig_flag== True:
#vutils.save_image(real_cpu,
# '%s/real_samples.png' % opt.outf,
# normalize=True)
#netG.eval()
#fake = netG(fixed_noise)
#cv2.imwrite(Save_pic_dir + str(i) +".jpg", mat)
#show the result
D_xf = errD_real_fuse.data.mean()
optimizerD.step()
save_out = output2
# train with fake
# if cv2.waitKey(12) & 0xFF == ord('q'):
# break
print('[%d/%d][%d/%d] Loss_D: %.4f Loss_G: %.4f D(x): %.4f D(G(z)): %.4f / %.4f'
% (epoch, 0, read_id, 0,
errD_real.data, D_x, 0, 0, 0))
if read_id % 20 == 0 and Visdom_flag == True:
plotter.plot( 'cLOSS', 'cLOSS', 'cLOSS', iteration_num, D_x.cpu().detach().numpy())
plotter.plot( 'cLOSS1', 'cLOSS1', 'cLOSS1', iteration_num, D_x1.cpu().detach().numpy())
plotter.plot( 'cLOSS12', 'cLOSS2', 'cLOSS2', iteration_num, D_x2.cpu().detach().numpy())
plotter.plot( 'cLOSS_f', 'cLOSSf', 'cLOSSf', iteration_num, D_xf.cpu().detach().numpy())
if read_id % 1 == 0 and Display_fig_flag== True:
#vutils.save_image(real_cpu,
# '%s/real_samples.png' % opt.outf,
# normalize=True)
#netG.eval()
#fake = netG(fixed_noise)
#cv2.imwrite(Save_pic_dir + str(i) +".jpg", mat)
#show the result
gray2 = (mydata_loader.input_image[0,0,:,:] *104)+104
show1 = gray2.astype(float)
#optimizerG.step()
#save_out = Fake
# train with fake
# if cv2.waitKey(12) & 0xFF == ord('q'):
# break
if validation_flag==False:
print('[%d/%d][%d/%d] Loss_D: %.4f Loss_G: %.4f D(x): %.4f D(G(z)): %.4f / %.4f'
% (epoch, 0, read_id, 0,
G_x, D_x, 0, 0, 0))
if read_id % 2 == 0 and Visdom_flag == True and validation_flag==False:
plotter.plot( 'l0', 'l0', 'l0', iteration_num, GANmodel.displayloss0.cpu().detach().numpy())
plotter.plot( 'l1', 'l1', 'l1', iteration_num, GANmodel.displayloss1.cpu().detach().numpy())
plotter.plot( 'l2', 'l2', 'l2', iteration_num, GANmodel.displayloss2.cpu().detach().numpy())
plotter.plot( 'l3', 'l3', 'l3', iteration_num, GANmodel.displayloss3.cpu().detach().numpy())
#plotter.plot( 'cLOSS1', 'cLOSS1', 'cLOSS1', iteration_num, D_x1.cpu().detach().numpy())
#plotter.plot( 'cLOSS12', 'cLOSS2', 'cLOSS2', iteration_num, D_x2.cpu().detach().numpy())
#plotter.plot( 'cLOSS_f', 'cLOSSf', 'cLOSSf', iteration_num, D_xf.cpu().detach().numpy())
if read_id % 1 == 0 and Display_fig_flag== True:
#vutils.save_image(real_cpu,
# '%s/real_samples.png' % opt.outf,
# normalize=True)
#netG.eval()
#fake = netG(fixed_noise)
#cv2.imwrite(Save_pic_dir + str(i) +".jpg", mat)