Experimental results on miniimagenet dataset about deepemd HOT 8 CLOSED

hello, are you changing any parts of the code? i have double-checked before uploading. how about downloading the code and run it again?

from deepemd.

Hello, I downloaded the code again and ran it again. But the result is still the same.

from deepemd.

how about the dataset?

from deepemd.

Thank you very much for your reply. I use miniimagenet dataset downloaded from other places. Will this have any impact? Has your dataset been processed in advance？Then I'll download your dataset for testing and look at the results.

from deepemd.

hi, you may refer to this issue. might be related.

from deepemd.

Thank you. I'll download the dataset and test it again.
In addition, I have another question. When I train the cub dataset, there will always be such errors. Have you encountered them? How did you solve them?
RuntimeError: cuda runtime error (59) : device-side assert triggered at /pytorch/aten/src/THCUNN/generic/ClassNLLCriterion.cu:110

from deepemd.

Hello, I've solved that problem. I used the dataset you provided, and the test results are similar to the data you provided. In addition, I have solved the problem of cub dataset. In datautils.py of dataloder:

elif args.dataset == 'cub':
args.num_class = 100

The value of num_class needs to be changed to 100.
Finally, can you provide the visualization program in the paper?
I am really desperate for the answer.
Thanks in advance for your help.

from deepemd.

hi , thank your pointing our the issue.
for the visualization program, it is not compatible with the released code. You may drop me an email and I can send you the raw code for reference. you may also write one on your own.
After you obtain the flow matrix from opencv or QPTH, you actually get a correspondence matrix. Then you only need to write a script to crop&paste patches based on it. Then you use normalized weights to control the brightness.

from deepemd.