search

microsoft / View-Adaptive-Neural-Networks-for-Skeleton-based-Human-Action-Recognition

View Adaptive Neural Networks for High Performance Skeleton-based Human Action Recognition
MIT License
127 stars 58 forks source link

HOW TO DO THE FUSION #4

Open saadbenda opened 4 years ago

saadbenda commented 4 years ago

Hi guys, thank you for your incredible work. I want to know how exactly and precisly you do the fusion between the rnn and the cnn. I want to use that same logic. If you can send me some code it will be even better. Thank You Again.

shuidongliu commented 4 years ago

Hi, you can use the following functions to get the fusion. va_rnn_results and va_cnn_results are the predictions of last fc layer before softmax layer.

def softmax(data): e = np.exp(data - np.amax(data, axis=-1, keepdims=True)) s = np.sum(e, axis=-1, keepdims=True) return e / s

def acc(va_rnn_results, va_cnn_results, label): pred = va_rnn_results + 4 * va_cnn_results pred_label = np.argmax(softmax(pred), axis= -1) total = ((label-pred_label)==0).sum()

return float(total) / label.shape[0] * 100