https://github.com/dk-liang/transcrowd
[SCIS 22] TransCrowd: Weakly-Supervised Crowd Counting with Transformers
https://github.com/dk-liang/transcrowd
Last synced: 6 months ago
JSON representation
[SCIS 22] TransCrowd: Weakly-Supervised Crowd Counting with Transformers
- Host: GitHub
- URL: https://github.com/dk-liang/transcrowd
- Owner: dk-liang
- License: mit
- Created: 2021-04-19T01:19:18.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2023-06-15T09:33:11.000Z (almost 2 years ago)
- Last Synced: 2024-10-16T18:18:05.961Z (7 months ago)
- Language: Python
- Homepage:
- Size: 177 KB
- Stars: 94
- Watchers: 3
- Forks: 17
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# TransCrowd
* An officical implementation of TransCrowd: Weakly-Supervised Crowd Counting with Transformers. To the best of our knowledge, this is the first work to adopt a pure Transformer for crowd counting research. We observe that the proposed TransCrowd can effectively extract the semantic crowd information by using the self-attention mechanism of Transformer.
* Paper [Link](https://arxiv.org/abs/2104.09116)
## Overview
# Environment
python >=3.6
pytorch >=1.5
opencv-python >=4.0
scipy >=1.4.0
h5py >=2.10
pillow >=7.0.0
imageio >=1.18
timm==0.1.30
# Datasets
- Download ShanghaiTech dataset from [Baidu-Disk](https://pan.baidu.com/s/15WJ-Mm_B_2lY90uBZbsLwA), passward:cjnx; or [Google-Drive](https://drive.google.com/file/d/1CkYppr_IqR1s6wi53l2gKoGqm7LkJ-Lc/view?usp=sharing)
- Download UCF-QNRF dataset from [here](https://www.crcv.ucf.edu/data/ucf-qnrf/)
- Download JHU-CROWD ++ dataset from [here](http://www.crowd-counting.com/)
- Download NWPU-CROWD dataset from [Baidu-Disk](https://pan.baidu.com/s/1VhFlS5row-ATReskMn5xTw), passward:3awa; or [Google-Drive](https://drive.google.com/file/d/1drjYZW7hp6bQI39u7ffPYwt4Kno9cLu8/view?usp=sharing)
# Prepare data
```
cd data
run python predataset_xx.py
```
“xx” means the dataset name, including sh, jhu, qnrf, and nwpu. You should change the dataset path.
Generate image file list:
```
run python make_npydata.py
```
# Training
**Training example:**
```
python train.py --dataset ShanghaiA --save_path ./save_file/ShanghaiA --batch_size 24 --model_type 'token'
python train.py --dataset ShanghaiA --save_path ./save_file/ShanghaiA --batch_size 24 --model_type 'gap'
```
Please utilize a single GPU with 24G memory or multiple GPU for training. On the other hand, you also can change the batch size.
# Testing
**Test example:**
Download the pretrained model from [Baidu-Disk](https://pan.baidu.com/s/1OJZmZfDGOuHCVMtJwrPHUw), passward:8a8n
```
python test.py --dataset ShanghaiA --pre model_best.pth --model_type 'gap'
```
# Reference
If you find this project is useful for your research, please cite:
```
@article{liang2022transcrowd,
title={TransCrowd: weakly-supervised crowd counting with transformers},
author={Liang, Dingkang and Chen, Xiwu and Xu, Wei and Zhou, Yu and Bai, Xiang},
journal={Science China Information Sciences},
volume={65},
number={6},
pages={1--14},
year={2022},
publisher={Springer}
}
```