img_np = np.asarray(img).reshape(
HEIGHT, WIDTH, 3)[:, :, ::-1] #.astype('uint8') # HWC, RGB
# import pdb; pdb.set_trace()
#plt.imshow(img_np)
#plt.show()
img_np = (img_np / 255).astype('float32')
x = trn_tfms(
img_np
)[np.newaxis,
...] # shape (210, 280, 3) -> (3, 210, 210) -> (1, 3, 210, 210)
x = Variable(T(x), requires_grad=False, volatile=True)
output = learner.model(x, None) # shape (1, 14)
pred_indicators = transform_range_output(to_np(output[0]),
desired_ranges,
output_ranges)
print("network raw output", output)
print("pred_indicators", pred_indicators)
indicators_formatted = format_indicators(pred_indicators)
print("indicators_formatted", indicators_formatted)
ground_truth = drive.read_indicators()
print("ground_truth", ground_truth)
drive.controller(indicators_formatted)
drive.update_visualizations(indicators_formatted, ground_truth)
drive.write(False) # Shared data read, and TORCS may continue
drive.wait_key(1)
# drive.controller(ground_truth)
# drive.write(False) # Shared data read, and TORCS may continue
# Test 8: visualization
@contextmanager
def context(*args, **kwds):
drive.setup_shared_memory()
drive.setup_opencv()
try:
yield None
finally:
drive.close_shared_memory()
drive.close_opencv()
with context() as _:
drive.pause(False) # TORCS may share images and ground truth
print("Controlling: ", drive.is_controlling())
drive.set_control(True)
print("Controlling: ", drive.is_controlling())
input("Press key to start...")
while True:
if drive.is_written():
ground_truth = drive.read_indicators()
print(ground_truth)
drive.controller(ground_truth)
drive.update_visualizations(ground_truth, ground_truth)
drive.write(False) # Shared data read, and TORCS may continue
drive.wait_key(1)