Inquiry regarding the yaw range of images in the MGL dataset

[wangerniuniu]

During my usage, I noticed that the yaw range of the images is defined as -360° to 360°, rather than the conventional 0° to 360° range. I am perplexed by this design decision and would like to understand the reasoning behind it.

Could you kindly explain why you chose to define the yaw angle within the range of -360° to 360°? Are there any specific technical or practical requirements that led to this setting? Furthermore, what implications does this variation have on data processing and analysis compared to the traditional 0° to 360° range?

[sarlinpe]

The yaw predicted by OrienterNet should be in [0, 360°]:

OrienterNet/maploc/models/voting.py

Lines 244 to 245 in 71e89eb

	angle_index = indices % num_rotations
	angle = angle_index * 360 / num_rotations

GT data may be stored with a different range, but it doesn't matter because it is shifted to the same one when computing the evaluation error:

OrienterNet/maploc/models/metrics.py

Lines 14 to 17 in 71e89eb

	def angle_error(t, t_gt):
	error = torch.abs(t % 360 - t_gt.to(t) % 360)
	error = torch.minimum(error, 360 - error)
	return error