Is it possible to inference on images about cdistnet

[simplify23]

I don't quite understand you mean. Do you mean to evaluate with other lmdb datasets ?

[saichandra1199]

@simplify23 It means like is there any inference script so that we can direct test the model on images rather than lmdb files

[simplify23]

@saichandra1199 our code support for testing by txt labels. You can set cfg.test.is_test_gt in Config file for test, some code need to be change.

[saichandra1199]

@simplify23 so you are saying that we need ground truth txt file for every image which needs to be tested. Is that right ?
Can't we get direct predictions over set of images given using a model which is trained
So what you're saying that cfg.test.is_test_gt = True and i've to give image_dir path consisting of images and a txt file which consists of image name and ground truth text seperated by space

[simplify23]

If you can't get real txt labels, you can also try this method. The prediction result will save as submit.txt,you just need empty txt or note other code.
You can read the eval.py line 180 for change.

gt_file = os.path.join(cfg.test.model_dir, 'gt.txt')
pred_file = os.path.join(cfg.test.model_dir, 'submit.txt')
name_file = os.path.join(cfg.test.model_dir, 'name.txt')
write_to_file(gt_file, gt_list)
write_to_file(pred_file, pred_list)
write_to_file(name_file, name_list)