#CNN on visual speech recognition ####0.94Training Accuracy with 0.51 Validation Accuracy of 20 class lip reading ##Category ####Using MIRACL dataset ##Model VGG16 with two fully connected layers on top. Fully connected layer has Dropout layer(0.5) after them and l2 regularizer with value 0.1 and 0.5 (result show both). Facing overtraining problem. ##Parameter Optimizer is SGD with learning rate 0.0005 and decay 1e-6, momentum=0.9. ##Result ###Accuracy Chart #####Show result in 80 epoch. #####Training accuracy will raise to 93% in the end for l2=0.1 #####Validation gets best result(51%) when epoch is 21 and l2=0.1.
-
Notifications
You must be signed in to change notification settings - Fork 12
staywithme23/CNN-for-visual-speech-recognition
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
CNN for visual speech recognition
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published