CSRNet+IADM for RGBD crowd counting

We take CSRNet as backbone to develop our framework on ShanghaiTechRGBD benchmark.

Prerequisites

We strongly recommend Anaconda as the environment.

Python: 2.7

PyTorch: 0.4.0

Preprocessing

Generation the ground-truth density maps for training (please edit the dataset root path in the script).

python RGBD_GT_generation.py

Make data path files and edit this file to change the path to your original datasets.

python make_json.py

Training

Edit this file for training CSRNet-based IADM model.

bash train.sh

Testing

Edit this file for testing models.

bash test.sh

You can also test our checkpoints released at here