search

yangze0930 / NTS-Net

This is a PyTorch implementation of the ECCV2018 paper "Learning to Navigate for Fine-grained Classification" (Ze Yang, Tiange Luo, Dong Wang, Zhiqiang Hu, Jun Gao, Liwei Wang).
MIT License
450 stars 118 forks source link

I have a question about testing! #7

Open Xiaoccer opened 5 years ago

Xiaoccer commented 5 years ago

Hi, i have a question about testing. Why the test images use center crop instead of scaling to 448x448 directly?

doublemanyu commented 5 years ago

hi, do you train the model successfully? I have a question that need your help. What does the term_width in core/util.py means?

Xiaoccer commented 5 years ago

hi, i think the term_width is the width of terminal, you can try to use command line to train the model or fix the term_width number.

doublemanyu commented 5 years ago

hi, i think the term_width is the width of terminal, you can try to use command line to train the model or fix the term_width number.

Hi, I fix the term_width=50, then run util.py successfully. And I try to train the model by command line. The error stills there. such as RuntimeError: Expected object of type torch.cuda.LongTensor but found type torch.cuda.IntTensor for argument #3 'index'. Do you meet the same problem?

ygean commented 5 years ago

@doublemanyu Hey, please show more error details for this ERROR, but I guess you need to change at here , just have a try.

yangze0930 commented 5 years ago

@Xiaoccer I do not want to change the data distribution of training images and testing images. So I use the cropping strategy.

RunshengZhu commented 4 years ago

@yangze0930 So actually the accuracy on cropped test set can not be used as final performance?