Stacked Pooling for Boosting Scale Invariance of Crowd Counting

PyTorch implementation of "Stacked Pooling for Boosting Scale Invariance of Crowd Counting" [ICASSP 2020].

@inproceedings{huang2020stacked,
  title={Stacked Pooling for Boosting Scale Invariance of Crowd Counting},
  author={Huang, Siyu and Li, Xi and Cheng, Zhi-Qi and Zhang, Zhongfei and Hauptmann, Alexander},
  booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing},
  pages={2578--2582},
  year={2020},
}

This code is implemented based on https://github.com/svishwa/crowdcount-mcnn

	ShanghaiTech-A	ShanghaiTech-B	WorldExpo'10
Vanilla Pooling	97.63	21.17	14.74
Stacked Pooling	93.98	18.73	12.92

Dependency

1. Python 2.7
2. PyTorch 0.4.0

Data Setup

1. Download ShanghaiTech Dataset from
   Dropbox: https://www.dropbox.com/s/fipgjqxl7uj8hd5/ShanghaiTech.zip?dl=0
   Baidu Disk: http://pan.baidu.com/s/1nuAYslz
2. Create Directory mkdir ./data/original/shanghaitech/
3. Save "part_A_final" under ./data/original/shanghaitech/
   Save "part_B_final" under ./data/original/shanghaitech/
4. cd ./data_preparation/
   Run create_gt_test_set_shtech.m in matlab to create ground truth files for test data
   Run create_training_set_shtech.m in matlab to create training and validataion set along with ground truth files

Train

1. To train Deep Net+vanilla pooling on ShanghaiTechA, edit configurations in train.py

   pool = pools[0] 

   To train Deep Net+stacked pooling on ShanghaiTechA, edit configurations in train.py

   pool = pools[1] 

2. Run python train.py respectively to start training

Test

1. Follow step 1 of Train to edit corresponding pool in test.py
2. Edit model_path in test.py using the best checkpoint on validation set (output by training process)
3. Run python test.py respectively to compare them!

Note

1. To try pooling methods (vanilla pooling, stacked pooling, and multi-kernel pooling) described in our paper:

   Edit pool in train.py and test.py

2. To evaluate on datasets (ShanghaiTechA, ShanghaiTechB) or backbone models (Base Net, Wide-Net, Deep-Net) described in our paper:

   Edit dataset_name or model in train.py and test.py