search

rizkiarm / LipNet

Keras implementation of 'LipNet: End-to-End Sentence-level Lipreading'
MIT License
638 stars 226 forks source link

Overfitting with pre-trained models? #60

Open Adam-Oudaimah opened 6 years ago

Adam-Oudaimah commented 6 years ago

I am testing the pre-trained model on random videos which are not from the training dataset. The prediction result is still as the dataset shape (i.e, command(4) + color(4) + preposition(4) + letter(25) + digit(10) + adverb(4)) and not like the video i entered. this is an overfitting case. what should i do not to get rid of the overfitting but just to change the shape of the predicted sentence?

Help please, I have only 2 days until my final project demo!!!!!!

chahatagarwal commented 4 years ago

I am testing the pre-trained model on random videos which are not from the training dataset. The prediction result is still as the dataset shape (i.e, command(4) + color(4) + preposition(4) + letter(25) + digit(10) + adverb(4)) and not like the video i entered. this is an overfitting case. what should i do not to get rid of the overfitting but just to change the shape of the predicted sentence?

Help please, I have only 2 days until my final project demo!!!!!!

Did you get any solution ? and why is this a specific dataset format ?