Batch Processing Camera Trap Videos (with Megadetector)

This repository contains a simple commandline tool for applying megadetector to camera trap video data without extracting the frames first. It is based on MegaDetectorV6 (corresponding to PyTorchWildlife version 1.1.0). Please note that this repository does not have the functionality to use a different detector.

Important

The script will extract the frames for each video to a temporary directory using the specified FPS and, once each video has been processed, remove them again. The results will be stored in a pickle file (named something like results_part_1.pkl).

Installation

Please make sure PytorchWildlife is installed following the official instructions here.

Usage

Example of how to use the commandline tool:

python batch_video_detection.py \
    --weights="weights/MDV6b-yolov9c.pt" \ # pretrained megadetector weights
    --data_path="path/to/videos" \ # path to videos (currently only looking for .mp4 files)
    --output_fps=4 \ # frames per second
    --batch_size=8 \ # num of videos in batch
    --save_every=100 \ # saves the results for every 100 videos

For more options run:

python batch_video_detection.py --help