search

weinman / cnn_lstm_ctc_ocr

Tensorflow-based CNN+LSTM trained with CTC-loss for OCR
GNU General Public License v3.0
498 stars 170 forks source link

convergence problem #4

Closed ghost closed 6 years ago

ghost commented 6 years ago

image I'm confused Why doesn't the loss function go down? Who can tell me? I am a beginner.

Govan111 commented 6 years ago

Do you solve the problem?I'm coping with the problem that is same with yours.

weinman commented 6 years ago

Are you using the mjsynth data framework and default hyperparameters (e.g., learning rate) provided or your own data, etc.?

Govan111 commented 6 years ago

Yes,but I encountered this problem when I changed the vgg model to another feature extraction model.The main problem confused me a lot is that the feature extraction network,including down sampling and the input size ofBLATM ,is the same as the vgg model but loss did't converge.