Figure 4.

(Left) DECOLLE setup for DvsGesture recognition. Learning was performed on the dataset provided with Amir et al. (2017) and consists of 11 gestures. The network consisted of three convolutional layers with max-pooling. A local classifier is attached to every layer and followed by dropout for regularization. DECOLLE is fed with 1 ms integer frames. (Right) Classification Error for the DvsGesture task during learning for all local errors associated with the convolutional layers . Shadings indicate standard deviation across runs (5 runs for C3D, 10 runs for DECOLLE).