Neural blur classifier

Pragmatically sorting through my home lab. A simple neural classifier, trained on acozma/imagenet-1k-rand_blur from Hugging Face.

To Run

Environment

To build and run the project, first initialize your environment.

PIP:

python -m venv .myvenv
source .myvenv/bin/activate
pip install -r requirements.txt

Note: it is highly recommended to use CUDA. Ensure you are installing the GPU-accelerated versions of PyTorch.

To train:

python classifier.py train --steps 1000

To run:

python classifier.py inf --source /path/to/images  # image directory to be sorted
                         --output /path/to/output  # sorted images are cp'd to here in `normal`, `blurred` subdirs
                         --threshold 0.8           # confidence threshold for classifying an image as blurry (range: 0 to 1. default: 0.8)

Contributing

• Thanks to @mlindgren for adding HEIC support