help me in improving accuracy

[balajib363]

I followed your ideas to train a model. After training when I test with my own data my predictions are not upto the mark.
Please share me your accuracy level.
print('Test accuracy:', 100*score[1])

[serengil]

You are asking this for which source file? Would you mention the file name directly?

[balajib363]

facial-expression-recognition.py
dont mind for not mentioning the file name.

[serengil]

I've got the following metrics
Test loss: 2.279457058505316
Test accuracy: 57.42546670713874

BTW, you can directly use the pre-trained weights stored in the path: https://github.com/serengil/tensorflow-101/tree/master/model . Set fit variable false in the main program and put all those files in same directory.

[BalajiB3663]

Great job. I deployed with your models, it as good accuracy. But for better understanding just want to check can we improve our accuracy by tuning some hyper parameter like increasing learning epochs or any better ideas.
Thanks for all your response 🙂

[serengil]

Actually, you should re-design the model (add new convolution, pooling layers). But it requires test and test again. Also, you should train with gpu because it lasts very long.

BTW, kaggle competition winner model got %34 accuracy. Here, we've already imporved that accuracy (%57). I mentioned it in related blog post: https://sefiks.com/2018/01/01/facial-expression-recognition-with-keras/

[BalajiB3663]

Nice job.
Thanking you for quick clarification on related work.

Please can close the issue.