I got ffmpeg compiled with OpenMAX encoding acceleration support on RPi--yay! Now my problem is, I'm working with a protocol that insists on specifying its required resolution. As soon as i specify
-vf scale=[W]:[H] it reverts to CPU encoding. Since ffmpeg/v4l2 (at least) only lets you capture at resolutions supported by the camera hardware, I'm stuck between a rock and a hard place.
Clearly I'm CPU constrained here, and GPU encoding is the only thing that makes what I'm trying to do feasible. Video quality is not a primary concern. Is there a way to preprocess the video frames with a fast (e.g. linear) scaler to the size I need, and then encode with my hardware acceleration? I'm trying to get only about 5fps on a ARMv6 (RPi model B+), but I need up to 1280x720 (possibly up to 1080p, but I'm uncertain).
Edit: To head off possible dispute, yes it needs to be ffmpeg, not avconv.