MegaDetectorV6 for batch video detection

[andrewzamp]

Search before asking

• I have searched the Pytorch-Wildlife issues and found no similar bug report.

Question

I'm using MDv6 without problems for batch image detection, however I need to apply it to a video dataset. From the Jupyter notebook demos I'm seeing that the only application in this sense is on single video to produce an annotated one with bounding boxes. Is there a way to run MDv6 on a video dataset the same way I'd do it on batches of images, with the objective of producing a Timelapse-ready JSON file? I couldn't find anything apart from the process_video() method which only takes a single video as input and a callback where single image detection is carried out on targte frames, and returns the annotated video.

Additional

No response

[zhmiao]

Hello @andrewzamp, thanks for asking, and sorry for the late reply! Currently there is no batch video processing options, because we need to generate batches of frames within each videos. Therefore, we recommend you start with single video streaming process for your video processing. I think the processing time won't be too inferior. Please let us know how it goes! Thanks!