Home

Awesome

Learning To Count Anything: Reference-less Class-agnostic Counting with Weak Supervision

**Project Page | Latest arXiv | Dataset **

Michael Hobley, Victor Adrian Prisacariu.

Active Vision Lab (AVL), University of Oxford.

Environment

We provide a environment.yml file to set up a conda environment:

git clone https://github.com/ActiveVisionLab/ABC123.git
cd ABC123
conda env create -f environment.yml

Dataset Download

MCAC

Dowload MCAC to precompute ground truth density maps for other resolutions, occlusion percentages, and gaussian standard deviations:

cd PATH/TO/MCAC/
python make_gaussian_maps.py  --occulsion_limit <desired_max_occlusion>  --crop_size 672 --img_size <desired_resolution> --gauss_constant <desired_gaussian_std>;

Trained Model Download

We provide example weights for our models trained on MCAC. Put this in ./checkpoints/.

Example Training

To train the counting network:

python main.py --config ABC123;

Example Testing

To test a trained model on MCAC:

python main.py --config ABC123test;

To test a trained model on MCAC-M1:

python main.py --config ABC123testM1;

Citation

@article{hobley2023abc,
    title={ABC Easy as 123: A Blind Counter for Exemplar-Free Multi-Class Class-agnostic Counting}, 
    author={Michael A. Hobley and Victor A. Prisacariu},
    journal={Proceedings of the European Conference on Computer Vision},
    year={2024},
}