Hello, friend!I used a pre-deployment term -- Visualize - ranks. Why did the test report a failure of query_flip and Gallery_flip about abd-net HOT 2 OPEN

File "train_multi.py", line 132, in main
performance, dm.return_testdataset_by_name(name),
File "E:\jpf\ABD-Net.next-master\ABD-Net.next-master\torchreid\data_manager.py", line 33, in return_testdataset_by_name
self.testdataset_dict[name]['query_flip'], self.testdataset_dict[name]['gallery_flip']
KeyError: 'query_flip'

from abd-net.

Hi friend.

I refer to the answer in our supplement https://openaccess.thecvf.com/content_ICCV_2019/supplemental/Chen_ABD-Net_Attentive_but_ICCV_2019_supplemental.pdf. (The last paragraph of the section2.)

I also attach it here:
During testing, the images are re-sized to 384 × 128 and augmented only with normalization. Both the original and the horizontally flipped images are fed into the model, and the final feature embedding is the average of the two outputs.

Thanks for your interest!

from abd-net.