Identifying animals in images from timelapse cam

Nature Talk
1

As part of my growing house project, I have a solar powered camera set up that takes a photo every 20 minutes. You can find the latest photo at hbs-cam.f76.eu. It allows me to see what’s going on and how the trees are growing.

The other day, by chance, I saw a bird on one of the pictures: Typical Shrikes (Genus Lanius)

I am curious to see what other animals are passing by. But I don’t have the time to check each and every picture manually. I am thinking about running Megadetector on my server (Intel Core i5-13500, 64GiB RAM, no GPU) and getting animal photos flagged in XMP sidecar files. As all photos are synced to my local computer using Synthing, I can then use Adobe Bridge to quickly find those that are flagged.

Is this a good approach? And, if MegaDetector, should I use the Microsoft version or the agentmorris fork?

2

I successfully installed MegaDetector as follows. This is the Microsoft version, which offers the newer model (v6 vs. v5). In the second line, I installed the CPU optimized PyTorch, which – as far as I understand it – utilized my CPU’s AVX2 (i5-13500).

python -m venv PytorchWildlife
./PytorchWildlife/bin/pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
./PytorchWildlife/bin/pip install PytorchWildlife
./PytorchWildlife/bin/pip install soundfile librosa
cd PytorchWildlife/
mkdir src
cd src
git clone https://github.com/microsoft/MegaDetector
cd MegaDetector/
../../bin/pip install -e .

While I do get quite a number of false positives and negatives, especially with night vision photos, MegaDetector is certainly very useful. It simply is not possible to do all the work manually.