Robust Class-Agnostic Counter (RCAC) (ECCV2022)

1. Install packages

conda create --name RCAC 
source activate RCAC
conda install pip
conda install python==3.6.6
pip install torch-1.0.0-cp36-cp36m-linux_x86_64.whl
pip install numpy==1.19.2 six==1.15.0 pillow==8.0.1
pip install torchvision==0.2.0
pip install pandas==1.1.3 python-dateutil==2.8.1 pytz==2020.1
pip install scipy==1.5.3 opencv-python==4.4.0.44
pip install opencv-python==4.4.0.44 cycler==0.10.0 kiwisolver==1.2.0 matplotlib==3.3.2 pyparsing==2.4.7
pip install bbox==0.9.2 pyquaternion==0.9.9
pip install PyWavelets==1.1.1 decorator==4.4.2 imageio==2.9.0 networkx==2.5 scikit-image==0.17.2 tifffile==2020.9.3
pip install boxx==0.9.2.28 fn==0.4.3 pprofile==2.0.5 py-heat==0.0.6 pyopengl==3.1.5 seaborn==0.11.0 snakeviz==2.1.0 tornado==6.1
pip install cached-property==1.5.2 h5py==3.1.0

2. Dataset download and model

download FSC-147 according to link

download model link password:xtva

3. Generate Edge

Use the pre-trained RCF model from link and generate edge images

or

Download the edge images we have already generated via pre-trained RCF model. link password:k13v

4. File system

- dataroot_FSC_147
  - images_384_VarV2
    - 2.jpg
    - ...
  - gt_density_map_adaptive_384_VarV2
    - 2.npy
    - ...
- dataroot_FSC_147_edge
  - 2.jpg
  - ...

command to obtain our result

python ./code/test.py model=RCAC_augnum0_test comment=analysis_times_RCAC_aug0 \
dataroot_FSC_147=/root/dataset/FSC-147 dataroot_FSC_147_edge=/root/dataset/FSC-147/edges_gen_by_RCF \
model_for_load=/root/RCAC/RCAC_20.94.pth