This is an unofficial implementation of the CrowdFormer architecture: https://arxiv.org/abs/2203.03768.
We train and evaluate on two datasets ShaghaiTech (SH) and the yet to be realeased CraneCounting (CC) dataset from the recently accepted IJCAI'23 paper: Long-term Monitoring of Bird Flocks in the Wild.
To preprocess the SH dataset, use the prep_SH folder, which also contains the dataloader in prep_SH/dataset.py.
To run on the SH dataset, use the main.py file.
Similar commands are used for CC dataset. (e.g.,
main_CC.py,Prep_CC/)