DAANet

The official implementation of "Density-assisted Adaptive Alignment Network for Crowd Flux Estimation".

Installation

• Clone this repo in the directory root/DAANet/src/

  cd root/DAANet/src
  git clone https://github.com/LeoHuang0511/DAANet.git

• Create and activate the envirnment

  cd root/DAANet/
  pip install virtualenv
  virturalenv DAANet_env
  source ./DAANet_env/bin/activate

• Install the dependencies (Python 3.8.10, PyTorch 2.1.2)

  pip install -r requirements.txt

Datasets

• SenseCrowd:
  1. Download SenseCrowd from Baidu disk.
  2. Download the original dataset form here and the lists of train/val/test sets at link1 or link2, and place them to each dataset folder, respectively.
• CroHD: Download CroHD from here.
• CARLA: Download CARLA from here.

Pretrained Weights

• SenseCrowd: The weights trained with Sensecrowd can be download from here
• CroHD: The weights trained with CroHD can be download from here
• CARLA: The weights trained with CARLA can be download from here

Preparation

Put the downloaded datasets in the directory root/datasets/, forming the folder structure like:

root
├──DAANet
│   └──src
├──exp
│   └──pretrained
│       ├──SensCrowd.pth
│       ├──CroHD.pth
│       └──CARLA.pth
└──datasets
    ├──SensCrowd
    ├──CroHD
    └──CARLA
     

Training

• Run the following command to train your own model:

  python train.py --DATASET SENSE --GPU_ID 0

• The checkpoints would be saved in root/DAANet/exp/.
• Replace --DATASET to HT21 or CARLA to change the training dataset.

Testing

• Run the following command to test the model pretrained on SenseCrowd:

  python test_SENSE.py --MODEL_PATH ../exp/pretrained/SenseCrowd.pth --GPU_ID 0

• Replace test_SENSE.py to test_HT21.py or test_CARLA.py to test the model trained on CroHD or CARLA.
• Change --MODEL_PATH to load other pretrained weights.