Return values from LoFTR Matcher

[m0ose]

I am having trouble figuring out what all of the return values are that come from calling

batch = {'image0': img0, 'image1': img1}

# Inference with LoFTR and get prediction
with torch.no_grad():
    matcher(batch)

Here is a dictionary of them.

dict_keys(['image0', 'image1', 'bs', 'hw0_i', 'hw1_i', 'hw0_c', 'hw1_c', 'hw0_f', 'hw1_f', 'conf_matrix', 'b_ids', 'i_ids', 'j_ids', 'gt_mask', 'm_bids', 'mkpts0_c', 'mkpts1_c', 'mconf', 'W', 'expec_f', 'mkpts0_f', 'mkpts1_f'])

Specifically what is the difference between mkpts0_f and mkpts0_c, or what is hw0_i?
Is there somewhere where these are described?

PS. I ask because this is such a cool technology and I have a feeling it is capable of more than the demo I ran.

[tcourat]

I guess mkpts0_f stands for keypoints at fine scale and mkpts0_c stands for keypoints at coarse scale . So you want to use mkpts0_f and mkpts1_f .