New requirement: iter_frames about bb_binary HOT 3 CLOSED

I would prefer option 2 with the FrameContainer as second return value.

I would prefer that too. That way the caller can always get the values he is interested in.

I also like you suggested enhancement, but you have to decide how useful it would be for the tracking.

You already had some basic code for iter_frames, right? Do you want to implement it?

from bb_binary.

Would the function simple yield raw Cap'N Proto frames? Or would they also be converted to
numpy arrays? What if you are interested in the cam_id which is stored only in the FrameContainer at the top level? Would this use-case be out of scope of the iter_frames function?
I personally prefer start till end like the python range function.

from bb_binary.

I personally prefer start till end like the python range function.

Agreed

Regarding the cam_id: I do not like the idea of dropping the information for a convenience method like iter_frames. So here are three suggestions:

1. Move cam_id from FrameContainer to Frame in the Schema, making it possible to store frames from different cameras in one FrameContainer.
2. yield the FrameContainer object as dictionary (without frames) as second return value along with the Frame object
3. Convert the Frame to a python dictionary and add the missing information from the FrameContainer.

Otherwise it could be argued, that if you need the cam_id or other information from FrameContainer, use iter_fnames or make several passes with the cam parameter of iter_frames.

I would prefer option 2 with the FrameContainer as second return value.
The calling code then would look something like:

from numpy.lib.recfunctions import append_fields
for frame, fc in repo.iter_frames(start, end):
    arr = convert_frame_to_numpy(frame)
    cam_col = np.repeat(fc.camId, arr.shape[0])
    arr_ext = append_fields(arr, 'cam_id', cam_col, usemask=False)
    # ... do something with arr_ext like stacking

We also could enhance convert_frame_to_numpy() to automatically add and fill columns like:

for frame, fc in repo.iter_frames(start, end):
    arr = convert_frame_to_numpy(frame, add_columns={'cam_id': fc.camId})

From my perspective this would be the most practical solution.

from bb_binary.