Is There an Opportunity to Combine the Skip Convolution and the Frame Exit Approach for Classification Applications?

There is an opportunity to combine the skip convolutions and the frame exit approach for classification applications, says Mo Soyo. And coming up with some sort of better representation which reflects the continuity of vido in space and time 'is a promising direction', he adds. But there are many ways that crontising residuals or using motion estimation inside the network can also increase the deficiency of the processe. These are all very relevant to quacom, because we have lot of applications like for x r, for autonos, driving,. We are dealing with high frame rate vido data and camera use cases. This is where we can use to reduce the power and and the computation