Exposing native computer vision building blocks and/or components, in a way that leverages native processing of video frames before sending the pre-processed video frames (and associated data) into the web app (either into the GPU or CPU), perhaps leveraging a library like Khronos' OpenVX. (In fact, one way this could be thought about is by working with Khronos to define WebVX.) Being able to leverage optimized platform capabilities for doing well-known basic algorithms (image pyramids, simple feature extraction, image conversion, blur, etc) could speed up in-app CV, and also allow some of the effects that might be done in the synchronous GPU case to be done faster. Like the WebRTC discussion, if we wanted to pursue this, we would want to do that elsewhere, but the scenario has been brought up multiple times, so we should document, summarize and record it it here.