pip install torchvision pip install matplotlib
Downloading: "https://download.pytorch.org/models/retinanet_resnet50_fpn_v2_coco-5905b1c5.pth" to /home/drew/.cache/torch/hub/checkpoints/retinanet_resnet50_fpn_v2_coco-5905b1c5.pth
self.font = core.getfont(
File "/code/py_env/wff/lib/python3.10/site-packages/PIL/ImageFont.py", line 274, in init self.font = core.getfont( OSError: cannot open resource
Couldn't use Arial
https://stackoverflow.com/questions/65141291/get-a-list-of-all-available-fonts-in-pil
https://medium.com/pythons-gurus/what-is-the-best-face-detector-ab650d8c1225
This suggests RetinaFace is very good but they claim it can miss on large faces, which exceed the size of the image causing partial occlusion. Since it can work on partial occulsion, it must be that it just wasn't trained on partial occlusion where it's at the edge of the frame ( or it doesn't pad to allow this )
https://www.codegenes.net/blog/pytorch-retinaface/
recent pytorch retinaface code
https://github.com/serengil/retinaface
this looked good.
https://github.com/serengil/deepface
this wraps their retinaface and allows for searching.
We'll want a different db to allow clustering.
this uses arcface to recognize then detect using retinaface.
pip install open_clip_torch
directions: https://huggingface.co/docs/hub/en/open_clip