https://github.com/huangyongbobo/eFreeNet

Remote Sensing Object Counting through Regression Ensembles and Learning to Rank
https://github.com/huangyongbobo/eFreeNet

Last synced: 7 days ago
JSON representation

Remote Sensing Object Counting through Regression Ensembles and Learning to Rank

• Host: GitHub
• URL: https://github.com/huangyongbobo/eFreeNet
• Owner: huangyongbobo
• Created: 2022-11-03T07:54:28.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2023-04-28T13:09:51.000Z (over 1 year ago)
• Last Synced: 2024-08-02T15:30:15.134Z (3 months ago)
• Language: Python
• Homepage:
• Size: 3.37 MB
• Stars: 4
• Watchers: 2
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# eFreeNet

This website provides a PyTorch implementation of eFreeNet. The repository contains source code for the paper entitled ***"Remote Sensing Object Counting through Regression Ensembles and Learning to Rank".***

## Overall Framework

![](https://github.com/huangyongbobo/eFateNet/blob/main/architecture.png)

## Dataset

* Download RSOC datasets from [here](https://github.com/gaoguangshuai/Counting-from-Sky-A-Large-scale-Dataset-for-Remote-Sensing-Object-Counting-and-A-Benchmark-Method). After downloading the RSOC-Building dataset, you need to use the `RSOC_building_preprocess.py` to generate ground truths.  

* Download VisDrone2019-People dataset from [here](https://drive.google.com/file/d/19gh-ZF-FpoTNNtVh_gScRc9pFlqvktpU/view?usp=sharing).  

* Download VisDrone2019-Vehicle dataset from [here](https://drive.google.com/file/d/12bCfAWEVurX6Z0RuAbegywkY7Z-UDU19/view?usp=sharing).  

## Visualization

We visualize the feature maps and heat maps of the last convolutional layers. Models 3 (the traditional global regression), 5 (the traditional global regression coupled with learning to rank), and 9 (eFreeNet) are used for visualization. The results are shown as follows. 

![](https://github.com/huangyongbobo/eFreeNet/blob/main/visualization.png)

## Environment

```

python: 3.7

pytorch: 1.4.0  

torchvision: 0.5.0

cuda: 9.2 

numpy: 1.19.4

```

## Code Structure

* `extend_sample`: code for data imbalance alleviation as well as augmentation. 

* `Dataset`: code for the dataloader, which returns images and ground truths. 

* `model`: code for building eFreeNet. For different numbers of learners, you need to modify the network architecture slightly. 

* `ranking_loss`: code for learning to rank. 

* `ambiguity_loss`: code for imposing the ambiguity constraint. 

* `train`: code for training. 

* `test`: code for evaluation. 

## Reference

This project is for research purpose only. If you find this project is useful for your research, please cite our paper: 

```

@ARTICLE{10102292,

  author={Huang, Yongbo and Jin, Yuanpei and Zhang, Liqiang and Liu, Yishu},

  journal={IEEE Transactions on Geoscience and Remote Sensing}, 

  title={Remote Sensing Object Counting Through Regression Ensembles and Learning to Rank}, 

  year={2023},

  volume={61},

  number={},

  pages={1-17},

  doi={10.1109/TGRS.2023.3266884}}

```