writer.add_scalars('loss/per_epoch', {'train_loss': train_epoch_loss / n}, epoch) train_loss = 0 train_epoch_loss = 0 data_loader_iter = iter(data_loader) for img, mask in data_loader_iter: solver.set_input(img, mask) train_loss = solver.optimize() train_epoch_loss += train_loss #print(train_loss) #print(iter_count) if (iter_count % 1000) == 0: t_img = solver.test_one_img_to_real(img) #print(t_img.shape) #print(1) * 2 - 1 writer.add_image('raw_image/raw_image', t_img[0] * 255, iter_count) t_mask = solver.test_one_img_to_real(mask) writer.add_image('raw_mask/raw_mask', t_mask[0] * 255, iter_count) #iter_img = solver.test_one_img(img) iter_img = solver.test_one_mask_to_real(img) #[0,] #print(iter_img.shape) iter_img = iter_img[0, ] #print(iter_img.shape) iter_img = iter_img.reshape(1, 512, 512) #print(iter_img.shape) writer.add_image('gen_mask/gen_mask', iter_img, iter_count) #print(train_loss)
h, w, c = src.shape img = np.array(src) dst = src.copy() dist = [] temp = np.array(src) temp = np.array(temp, np.float32).transpose(2,0,1)/255.0 dist.append(temp) dist = np.array(dist) img = torch.Tensor(dist) solver.set_input(img)#, mask) _ = solver.forward()#optimize() t_img = solver.test_one_img_to_real(img) t_img= t_img[0] t_img = t_img.transpose(1, 2, 0) t_img_out = t_img t_img_out = t_img_out.transpose(2,0,1) t_img_out = np.uint8(t_img_out.transpose(2,0,1)) #print(t_img_out.shape) iter_img = solver.test_one_mask_to_real(img)#[0,] iter_img = np.uint8(iter_img) #Image.fromarray(t_img_out).save(RESULT_DIR + 'raw_' + str(iter_count) + '_' + str(j) + '.jpg') t_out = Image.fromarray(iter_img, mode='L') #t_out.save(RESULT_DIR + 'result_' + str(iter_count) + '_' + str(j) + '.jpg')