loss = MyLoss(images, predictions)
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
train_loss(loss)
print("TENSORBOARD...")
tb = program.TensorBoard()
tb.configure(argv=[None, '--logdir', log_dir])
url = tb.launch()
batch_size = 8
num_train_steps = 1000000
for step in range(num_train_steps):
batch = reader.GetNextBatch(batch_size)
images = Preprocessor.preprocess_image_pairs(batch)
if (step % 10 == 0):
# Show first pair:
im1 = images[0, :, :, :3]
im2 = images[0, :, :, 3:]
mean = Preprocessor.mean
mean = np.squeeze(mean, axis=0)
img_to_show_1 = im1 + mean
img_to_show_2 = im2 + mean
cv2.imshow('image 1', img_to_show_1)
cv2.imshow('image 2', img_to_show_2)
# Run network on first pair:
imgs_to_predict = np.expand_dims(images[0, :, :, :], axis=0)
flows, motions, depths, ego = model(imgs_to_predict, training=False)
