italian_fingerspelling_recognition

Dataset + convolutional neural network for recognizing Italian Sign Language (LIS) fingerspelling gestures.
CNN with Keras for recognizing letters of the alphabet in LIS (22 classes).
Fore more info see documentation.pdf

Folder structure

dataset\
...... dataset1\ dataset for training
...... dataset2\ dataset for training
...... dataset3\ dataset for training
...... dataset1-signer-independent-test\ dataset for testing
...... dataset2-signer-independent-test\ dataset for testing
...... dataset3-signer-independent-test\ dataset for testing
plots\ accuracy and loss plots
weights\ weights for each dataset/training
statistics\ txt with statistics about accuracy and loss during training
cnn_model.py base model
cnn_model_batch_normalization.py model with batch normalization
train_cnn.py model training
test_cnn.py model testing (confusion matrix and predict)

This repo contains weights, statistics and plots for the base model trained on dataset 1 (250 epochs), dataset 2 (150 epochs) and dataset 3 (50 epochs).

Dataset

The complete dataset is downloadable at the following link: https://drive.google.com/file/d/1AFcb2VnGCn2OslIlB6kFpVDAS3PMNqvs/view?usp=sharing

Credits

Margherita Donnici & Gianluca Monica