About Res50 accuracy on ShanghaiTechB

[ckcraig01]

Dear Author, thank you fro providing this awesome framework. I have trained the Res50 model with ShangHaiTechB dataset but only get MAE with 8.4 instead of 7.7 (reported), is there any hyperparameter setting that required?

[haoranD]

Thanks for your amazing working @gjy3035. I also got a result which around MAE 8.4. Could you please give us any suggestion?

Thanks for your question. @ckcraig01 I got a quite similar result to you. Could you please give me any suggestion?

Much appreciated.

[gjy3035]

maybe compare the setting.py and config.py in the results reports folder.

[ckcraig01]

Thank you author, and @haoranD thanks. I have reviewed the setting.py/config.py, found that the differences are only on the size of val batch and the training epoch. but from "12-28_09-18_SHHB_Res50_1e-05.txt" I found from the log, the training can reach 7.7MAE in 200 epochs. So @haoranD please let me know if you have other discover

[gjy3035]

I upload the code that achieves MAE of 7.7, you can run it. Note that it is a python2 version.

code.zip

[haoranD]

Cool, working on it. Cheers