AdaCrowd: Unlabeled Scene Adaptation for Crowd Counting

This repository contains the code for AdaCrowd: Unlabeled Scene Adaptation for Crowd Counting
IEEE Transactions on Multimedia 2021

Setup

Create a python virtual environment using either virtualenv or conda.

Install all the dependencies using setup.py

python setup.py install

Datasets

The details related to all the crowd counting datasets can be found in the following links.

1. WorldExpo'10
2. Mall
3. PETS
4. FDST
5. CityUHK-X

Note: For processing datasets, please refer to C-3-Framework. Also update DATA_PATH in the setting.py for a particular dataset before using it for either training/testing.

Training

1. Baseline Methods

   Refer to config/baselines.py configuration file to train baseline methods using different backbone networks (e.g. VGG-16 / ResNet-101) and to make appropriate changes.

   python train_baselines.py

2. AdaCrowd Methods Refer to config/adacrowd.py configuration file to train adacrowd methods for different backbone networks.

   python train_adacrowd.py

Testing

Update MODEL_PATH in either config/adacrowd.py or config/baselines.py to test the model.

Refer to the following command-line arguments to run test_adacrowd.py.

optional arguments:
  -h, --help            show this help message and exit
  -mn MODEL_NAME, --model_name MODEL_NAME
                        Name of the model to be evaluated
  -mp MODEL_PATH, --model_path MODEL_PATH
                        Path of the pre-trained model
  -data {WE,UCSD,Mall,PETS,City}, --dataset {WE,UCSD,Mall,PETS,City}
                        Dataset to be evaluated
  -k K_SHOT, --k_shot K_SHOT
                        Number of K images used for computing the affine transformation parameters
  -norm NUM_NORM, --num_norm NUM_NORM
                        Number of normalization layers
  -gpu GPU_ID, --gpu_id GPU_ID
                        GPU ID
  -r RESULT_FOLDER, --result_folder RESULT_FOLDER
                        Path of the results folder
  -t TRAILS, --trails TRAILS
                        Number of random trails to calculate mean and std scores

Similarly, refer to the following command-line arguments to run test_baselines.py.

optional arguments:
  -h, --help            show this help message and exit
  -mn MODEL_NAME, --model_name MODEL_NAME
                        Name of the model to be evaluated
  -mp MODEL_PATH, --model_path MODEL_PATH
                        Path of the pre-trained model
  -data {WE,UCSD,Mall,PETS,City}, --dataset {WE,UCSD,Mall,PETS,City}
                        Dataset to be evaluated
  -norm NUM_NORM, --num_norm NUM_NORM
                        Number of normalization layers
  -gpu GPU_ID, --gpu_id GPU_ID
                        GPU ID
  -r RESULT_FOLDER, --result_folder RESULT_FOLDER
                        Path of the results folder

Citation

@article{reddy2020adacrowd,
  title     =   {AdaCrowd: Unlabeled Scene Adaptation for Crowd Counting},
  author    =   {Reddy, Mahesh Kumar Krishna and Rochan, Mrigank and Lu, Yiwei and Wang, Yang},
  journal   =   {arXiv preprint arXiv:2010.12141},
  year      =   {2020}
}

Acknowledgements

This project borrows code from C-3-Framework.

License

The project is licensed under MIT license (please refer to LICENSE.txt for more details).