Adamas

the code for evaluation in Photo2label of datasets-cityscapes might be wrong?

1

hi i notice that your code convert the color-photo into 14-class label but the output of net has 34 class ,i think this might convert some pixel to the wrong...

issue when command :python training_script.py --batchsize 2 -- dataset_name IWSLT --language_direction G2E

3

downloading de-en.tgz File "training_script.py", line 103, in train_transformer train_token_ids_loader, val_token_ids_loader, src_field_processor, trg_field_processor = get_data_loaders( tarfile.ReadError: not a gzip file

Rotation

3

Nice work, but I visualized the matching of Hpatches and superpoint failed in all rotation case.maybe rotation not include in training augmentation?

my loss always vary around 17 ,i have tried change the lr but it does not work ,could you please help me out

Rubustness against Rotation

1

seem like that the pretrained model privided is not rubust agsinst Rotation,Right？

baiduNetdisk link for the full dataset

6

here is the baiduNetdisk link for the full dataset: https://blog.csdn.net/sinat_41840241/article/details/113788215

Hi i run your command provided in readme of new version for 500k step but the psnr reached only to 28 ,far away from your metrics (30.8) in your paper...

test In Casia

5

Hi there thanks for your sharing have you test your model in datasets CASIA

effect not so good

1

I run the file demo.ipynb but the output is not as good as the original one posted in the web