retrain

[Lucifer1002]

Hello, I would like to ask how to retrain on the basis of the pre-trained model to enhance the network's resistance to rotation？How to implement？

[jiangwei221]

Hi, in order to obtain the correct correspondences, you need to rotate the camera pose, also the image and depth.
You can check:

COTR/COTR/cameras/capture.py

Lines 407 to 418 in bcc9ce7

	def rotate_capture(cap, rot):
	if rot == 0:
	return copy.deepcopy(cap)
	else:
	rot_pose = rotate_camera_pose(cap.cam_pose, rot)
	rot_cap = copy.deepcopy(cap)
	rot_cap.cam_pose = rot_pose
	if hasattr(rot_cap, 'captured_image'):
	rot_cap.captured_image.rotation = rot
	if hasattr(rot_cap, 'captured_depth'):
	rot_cap.captured_depth.rotation = rot
	return rot_cap

However, this implementation will cause black borders which can be a problem...