search

megvii-research / HomoGAN

This is the official implementation of HomoGAN, CVPR2022
43 stars 3 forks source link

Query about train and evaluation #9

Open florallam opened 1 year ago

florallam commented 1 year ago

Hello! Thank you for sharing your code, I do have a query about the training and evaluation portion of the code.

In the pre-training phase, there is a train_and_evaluate() function which would call the compute_eval_results() method. This would require the data in evaluate to contain the pt_set and matche pts, pt_set = list(map(eval, data_batch["pt_set"])) pt_set = list(map(lambda x: x['matche_pts'], pt_set))

which are not present in the evaluation dataset data dictionary data_dict = {"imgs_patch_rgb": imgs_patch_rgb, "imgs_gray_full": imgs_gray_full, "imgs_full": imgs_full, "imgs_gray_patch": imgs_gray_patch, "ori_size": ori_size, "img_name": img_names[0], "pts": pts, "start": start}

Is the evaluation method supposed to have pt_set and matche_pts or a different loss metric should be considered? Thank you!

one123455 commented 2 months ago

I have encountered the same problem. Have you resolved it？