Didn't you share the trained models?

[zhouhao-learning]

Hello, I'm glad to see your code, but when I execute python video-example.py', the probability of predicting each expression is almost the same. There is no obvious difference. Like the results shown below, it seems that the probability of happy'should be greater, but the result is not the same:

So I think it may be that your public model is not well trained. I would like to ask if you can publish your trained model?

[JustinShenk]

Hi Kyrie, Thanks for your message. I have seen this error caused by insufficient preprocessing, rather than issues with the model. Can you maybe share the video clip with me for testing? shenkjustin@gmail.com. I will take a look at it and let you know when it is fixed. Justin

…

On Wed 17. Jul 2019 at 07:47, Kyrie ***@***.***> wrote: Hello, I'm glad to see your code, but when I execute python video-example.py', the probability of predicting each expression is almost the same. There is no obvious difference. Like the results shown below, it seems that the probability of happy'should be greater, but the result is not the same: So I think it may be that your public model is not well trained. I would like to ask if you can publish your trained model? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACOLMZBQVM2LT5BCQ3LMQW3P72W6HANCNFSM4IEMLYIQ> .

[zhouhao-learning]

@JustinShenk
Ok, I have sent you the test video, please check it.

[zhouhao-learning]

@JustinShenk
The image prediction results I got from running example.py are also obviously wrong, so I suspect that it should be a model problem, not quite a problem with data preprocessing.

[JustinShenk]

The issue with the video was due to nesting TensorFlow sessions (not recommended). Fixed in f1322de. Please try again and let me know how it goes.

[JustinShenk]

Example:

[zhouhao-learning]

@JustinShenk
Ok, thank you very much for your reply.After you modify it, the effect is obviously correct.