Sign-Language-Digits-Dataset

Dataset used : https://github.com/ardamavi/Sign-Language-Digits-Dataset

Image size: 100x100

Color space: RGB

Sample Images:

0 1 2 3 4 5 6 7 8 9

Softwares and libraries used : Python, Tensorflow, keras, Sklearn.

Models used for Transfer Learning:

VGG16
RESNET50

VGG16 model:

Built the model using Keras Library.
Fine tuned the last classification layer for 10 classes and used imagenet weights for training.
Training Data : 80% of Data
Testing Data  : 20% of Data
Batch Size = 32 Epoch = 50
Training accuracy : 99.8%
Testing Accuracy  : 92.3%

RESNET-50 model

  Built the model using Keras Library.
  Fine Tuned the last classification layer for 10 classes and used the imagenet weights for training.
  Training Data : 80% of Data
  Testing Data  : 20% of Data
  Batch Size = 64 Epochs = 50
  Training accuracy : 99.8%
  Testing Accuracy  : 80.5%
  
  Batch Size :32 Epochs = 50
  Training Accuracy : 99.15%
  Testing Accuracy : 65.15%

Currently trying to make more data for each class using the following Data Augmentation Techniques

  Scaling
  Rotation (at 90 degrees)
  Rotation (at finer angles)
  Flipping
  Adding Salt and Pepper noise
  Lighting condition