transform_trans_out about timecycle HOT 1 OPEN

I was looking at the transform_trans_out in model_simple.py function and I noticed that the 2D rotation matrix is multiplied by 1/3. Could you please help figure out the reason for that? Shouldn't the 2D matrix be correct as is?

TimeCycle/models/videos/model_simple.py

Lines 148 to 151 in 16d33ac

	trans_out1_0 = 1.0 / 3.0 * torch.cos(trans_out1_theta).unsqueeze(1)
	trans_out1_1 = - 1.0 / 3.0 * torch.sin(trans_out1_theta).unsqueeze(1)
	trans_out1_3 = 1.0 / 3.0 * torch.sin(trans_out1_theta).unsqueeze(1)
	trans_out1_4 = 1.0 / 3.0 * torch.cos(trans_out1_theta).unsqueeze(1)

Hi Abdalla, I am looking at this code too. I guess this is because the tracked patch size is one-third of the original image size. When transformation theta is fed to the sampling function 'GeometricTnf', you shall expect to get a patch that is one-third of the size of the image. Multiplied by one-third, it does not change the rotation matrix but changes the sampling scale in fact. So that means if you wish to change the size of the sampled patch, the coefficient 1.0/3.0 should be changed accordingly

from timecycle.