Haozhe-Xing/SARNet

The trained weights do not reproduce the results.

yy13138 opened this issue · 8 comments

I'm trying to train on CAMO with 1000 images, hoping to get matching test results, but on the test set (250 images), my Smeasure is always around 40, which is 40% lower and faster than the one in the paper. Is it my training method?

the dataset you are using maybe not right, the train datasets contains CAMO and COD10K dataset, the number of the whole train dataset is about 4000.

Thank you. I'll reorganize right away. You're so kind and timely.

you are welcomed, thanks for your attention on our project.

COD10K-v3 dataset The training set has 6000 images, are you using this? What about other versions? Thank you for your reply. Thank you.

COD10K-v3 dataset The training set has 6000 images, are you using this? What about other versions? Thank you for your reply. Thank you.

I used the previous version, thanks for the reminder, I will update the dataset to git when I have time.

you can use the new version to train your model

First of all, thank you for your kind reply. I have found the required data set according to your tips. Thank you again. Here is the dataset address, I think anyone who needs data will also follow the URL to find the correct data, wish you good luck in your research.

https://drive.google.com/file/d/1D9bf1KeeCJsxxri6d2qAC7z6O1X_fxpt/view