ValueError: Unable to configure handler 'stderr': bad argument type for built-in operation

[lihao056]

I follow the README.md，but i can't run the train.py and
the error is ValueError: Unable to configure handler 'stderr': bad argument type for built-in operation

can you tell me how to use it? thank

[lucasb-eyer]

Can you try changing this line from common.ColorStreamHandler to logging.StreamHandler?

I fear this is a windows-problem with the colored logger; the workaround would disable colors in output, but I really would prefer if someone could help us finding a fix. It's probably a wrong argument type being passed into a ctypes function somewhere in the _WinColorStreamHandler.__init__ here.

What version of Python are you using?

[lihao056]

thanks for you answer, I can run the code.
my python version is 3.5.2 and i use windows10
due to the lack of GPU, I decrease the batch_size and the accuracy is that

it is right?

[Pandoro]

@celticssssss judging from your screenshot you trained on Market-1501 for which your scores, when you compare them to our results, are obviously pretty bad. Without any details about your exact setup we can't judge if it is "right", but we also never did exhaustive experiments w.r.t. far smaller batch sizes.

In any case your scores are no where near where they should be on Market-1501, however since your GPU only has 3GB of memory you probably decreased your batch size by at least a factor of 3 which could very well be a reason why you get low scores.

[lucasb-eyer]

Also, please be more specific if you expect help. "I decrease the batch_size" can mean anything. But yes, with too small of a batch size, you are basically not mining semi-hard samples anymore and I would expect the score to become worse.