For my impulse, I chose images of 160x160 and used Edge Impulse's FOMO. One of the neat features of the Edge Impulse Python SDK is that you can split a video frame into 2 frames (left and right), so the frame actually becomes two 160x160 images. This worked well for a nice view of my left and right hands. I could then run both of those images in the classifier. If I was just classifying only one image, since an equal dimension image is required for FOMO, the image would have been squeezed and not able to see both hands.