Transfer Learning to Predict Sign to Text from a Webcam. For Humans.

For each letter, you must reproduce the sign appearing on the sign alphabet (see the correspondence table at the bottom) and long press on the corresponding box
(on the right) so that it is recorded at different locations of the selected box. You must do it again for all the letters.

Then you have to press Train Model and finally Predict to make it work.

(Click on the 'A' Sample to add the current camera view as an example for that control.)

TRAIN MODEL PREDICT

Learning rate

0.00001 0.0001 0.001 0.003

Batch size

0.05 0.1 0.4 1

Epochs

10 20 40

Hidden units

10 100 200

Sign :

Loading the model...

'A' Sample

0 A examples

'B' Sample

0 B examples

'C' Sample

0 C examples

'D' Sample

0 D examples

'E' Sample

0 E examples

'F' Sample

0 F examples

'G' Sample

0 G examples

'H' Sample

0 H examples

'I' Sample

0 I examples

'J' Sample

0 J examples

'K' Sample

0 K examples

'L' Sample

0 L examples

'M' Sample

0 M examples

'N' Sample

0 N examples

'O' Sample

0 O examples

'P' Sample

0 P examples

'Q' Sample

0 Q examples

'R' Sample

0 R examples

'S' Sample

0 S examples

'T' Sample

0 T examples

'U' Sample

0 U examples

'V' Sample

0 V examples

'W' Sample

0 W examples

'X' Sample

0 X examples

'Y' Sample

0 Y examples

'Z' Sample

0 Z examples

At Montreal.AI, we believe that artificial intelligence must be ethical, humanistic and developed to reduce social inequalities.
We are delighted to present this sign language translated into text. If you have ideas for #AI4Good projects, please let us know.

This pre-alpha demo version supports the 26 letters of the English alphabet. We use a pretrained MobileNet model and train another model
using an internal mobilenet activation to predict 26 different classes. Thank you TensorFlow.js for your flexible and intuitive APIs.



A Pure JavaScript Implementation Built On Top Of : PoseNet And Transfer Learning Via Webcam.
An #AI4Good Deep Learning Application By : @Montreal_AI