Accuracy of the RNN for quickdraw Classification

[zhiweige]

Hello, Thanks for sharing the work on quickdraw classification. I have tried to train the network using the default settings in the train_model.py (I changed the --steps=1000000), and I got the accuracy about 55%. I wonder whether I missed something during the training? Thanks.

[zhiweige]

It's the accuracy curve during the training. Plz check.

And I also noticed that the loss for eval increased a lot. It seems overfitting happened.

[bitfort]

Thanks for reaching out, adding someone who can help :)

[lukaszkaiser]

Sorry, I don't know much about this code.

[zhiweige]

I noticed that the RNN uses biLSTM in tf.contrib, and the biLSTM in tf.contrib doesn't support the seq processing well. I don't know if these factors have relationship for the poor performance of the CNN+RNN classification network.

[bassmaamn]

Hello, can you share the code of how you displayed accuracy and loss in tensorboard? thank you

[michaelhuang74]

Hi bassmaamn,

You need to launch the tensorboard to see the graphs. Once the training is done, there is a file event.out.tfevents.... along with the check point files. Then you launch tensorboard as: tensorboard --logdir /path-to-save-event-file After tensorboard is launched, it will give you a port number with host name. You can see the visualization using web browser.

[tensorflowbutler]

Hi There, We are checking to see if you still need help on this, as this seems to be considerably old issue. Please update this issue with the latest information, code snippet to reproduce your issue and error you are seeing. If we don't hear from you in the next 7 days, this issue will be closed automatically. If you don't need help on this issue any more, please consider closing this.