qinnzou

I have uploaded all the pretrained models to BaiduYun. The download link is given in the Github site. You can also find it from https://pan.baidu.com/s/1u9ABAPouZL2nIMT4E7uAQw passcodes：w30b -----原始邮件----- 发件人:asharalam11 发送时间:2020-05-09 16:16:51...

The releaased code is running on Pytorch0.4.0. If adapted to Pytorch1.1.0 or above, `optimizer.step()' should be before `lr_scheduler.step()'. For the problem of out of memory, you may have to set...

Do you get the results?

We compute the accuracy, precision, recall, f1 by comparing the detected results and the ground truth. The codes can be found in the function `evaluate_model' in the test.py file.

Check the val_index.txt. It corresponses to the test samples. Contact me if there's problem. Thanks.

train_index.txt and val_index.txt include the data of step 1, 2 and 3. train_index_new.txt and val_index_new.txt only contain the data of step 1.

train_index.txt and val_index.txt have been uploaded @xxzcpsmemeda @hellosher

An nvidia 1060 or up gpu may satisfy the requirement.

If you use the pretrained model, the test results will be one pixel wide. If the crack ground truth are with width more than 1px, then you can trainyour own...

We do not use the socket in our codes. Is it the problem when you are using cloud computing resources?