Skip to content

Single Image Crowd Counting (CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting)

License

Notifications You must be signed in to change notification settings

svishwa/crowdcount-cascaded-mtl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting (Single Image Crowd Counting)

This is implementation of the paper CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting for single image crowd counting which is accepted at AVSS 2017

Installation

  1. Install pytorch

  2. Clone this repository

git clone https://github.com/svishwa/crowdcount-cascaded-mtl.git

We'll call the directory that you cloned crowdcount-cascaded-mtl ROOT

Data Setup

  1. Download ShanghaiTech Dataset

    Dropbox: https://www.dropbox.com/s/fipgjqxl7uj8hd5/ShanghaiTech.zip?dl=0

    Baidu Disk: http://pan.baidu.com/s/1nuAYslz

  2. Create Directory

mkdir ROOT/data/original/shanghaitech/  
  1. Save "part_A_final" under ROOT/data/original/shanghaitech/

  2. Save "part_B_final" under ROOT/data/original/shanghaitech/

  3. cd ROOT/data_preparation/

    run create_gt_test_set_shtech.m in matlab to create ground truth files for test data

  4. cd ROOT/data_preparation/

    run create_training_set_shtech.m in matlab to create training and validataion set along with ground truth files

Test

  1. Follow steps 1,2,3,4 and 5 from Data Setup

  2. Download pre-trained model files:

    [Shanghai Tech A]

    [Shanghai Tech B]

    Save the model files under ROOT/final_models

  3. Run test.py

    a. Set save_output = True to save output density maps

    b. Errors are saved in output directory

Training

  1. Follow steps 1,2,3,4 and 6 from Data Setup
  2. Run train.py

Training with TensorBoard

With the aid of Crayon, we can access the visualisation power of TensorBoard for any deep learning framework.

To use the TensorBoard, install Crayon (https://github.com/torrvision/crayon) and set use_tensorboard = True in ROOT/train.py.

Other notes

  1. During training, the best model is chosen using error on the validation set.

  2. 10% of the training set is set aside for validation. The validation set is chosen randomly.

  3. Following are the results on Shanghai Tech A and B dataset:

             |     |  MAE  |   MSE  |
             ------------------------
             | A   |  101  |   148  |
             ------------------------
             | B   |   17  |    29  |
    

    It may be noted that the results are slightly different from the paper. This is due to a few implementation differences as the earlier implementation was in torch-lua. Contact me if torch models (that were used for the paper) are required.

About

Single Image Crowd Counting (CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published