Feature Question: Frame analysis with cv.VideoCapture().read() instead of VideoReader

[JensBlack]

Hi,

i love your code and project and wanted to test out the implementation of deepposekit generated models in my own pipeline (DeepLabStream) . Unfortunately I am already working with the "base" class VideoCapture from OpenCV and cannot simply implement your VideoReader into my existing framework (as far as I can tell).

While your code suggests that any model exported and loaded needs the VideoReader for prediction, I assume it is a function that will pass the frames directly, with the "pure" predict() function somewhere deep within. Even though batch-size = 1 is possible to set and the VideoReader is only passing frames without altering them (if gray = False), I was not able to pass frames directly to model.predict(). I did not find the code for your model.predict() yet. When passing a single frame from VideoCapture.read() to model.predict(frame), I get a dimension error (I can post it in more detail if you'd like) which I assume means that simply passing it a frame does not work.

So my question is: Is there a way to directly access the model.predict function within each loaded model and pass it a single frame.

BTW I am using your (fly) data repository for testing now, thank you for providing it!

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.

[JensBlack]

found a solution myself.