Train a Neural Network to accurately classify Rock, Paper, Scissors, Lizard and Spock. Our Neural Netwrok are based on mobilnet and you just need to train the customized last few layers. So you only need to collect a small number of samples to train a good model.
Collecting Data: : Collecting between 50 to100 images for each hand gesture should be good enough for most models. However, you can experiment and try to train difference model with less or more images. Place a specific gesture in front of the computer's camera and click the corresponding button of your gestures. You will see how many different categories of images are currently collected. The example gestures are shown below. When collecting data, it is better to move gestures to different positions in front of the camera, as this helps the generalization ability of the model.
Training: Press the Button Train Network, You can see the training process on the right screen. It lasts about 2 minuts according to the amount of training data you collected and the configuration of your computer. There is a alert will be poped when training is finished.
Start Predicting: Take a gesture in front of camera and the Neural Network you have trained will give you a guess.
© 2020 zhangruochi.com