CrowdRec

KO | EN

gitlite — search

#javascript #python #llm #golang #machine-learning #rust #deep-learning #linux #java #awesome #kubernetes #awesome-list #pytorch #openai

(★ 10)

The code for the 1st place solution of 2022 GigaCrowd challenge.

CrowdRec 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# CrowdRec

The code for the 1st place solution of [2022 GigaCrowd challenge](https://gigavision.cn/track/track/?nav=GigaCrowd&type=nav&t=1678150678599).

> CrowdRec: 3D Crowd Reconstruction from Single Color Images 
> [Buzhen Huang](http://www.buzhenhuang.com/), [Jingyi Ju](https://me-ditto.github.io/), [Yangang Wang](https://www.yangangwang.com/#me) 
> \[[Paper](https://arxiv.org/abs/2310.06332)\]

![figure](https://raw.githubusercontent.com/boycehbz/CrowdRec/main/assets/teaser.jpg)

The code is tested on Windows 10 and WSL2 (Windows Subsystem for Linux) with an NVIDIA GeForce RTX 3090 GPU and 64GB memory.

## Demo

### Installation
Create conda environment and install dependencies.
```
conda create -n crowdrec python=3.8
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu111 # install pytorch
pip install -r requirements.txt
```
Download pycocotools from [cocoapi](https://github.com/philferriere/cocoapi) for win10, and install this package:
```
cd/d cocoapi/PythonAPI
python setup.py build_ext --inplace
python setup.py build_ext install
```
Due to the licenses, please download SMPL model file [here](http://smplify.is.tuebingen.mpg.de/).

Download the pretrain models from [pretrain_model](https://pan.baidu.com/s/1-_K3W_ImI3ESl4uygtKV-w?pwd=06le).

Finally put these data following the structure as below:
```
${ROOT}
|---data
    |---SMPL_NEUTRAL.pkl
    |---...
|---pretrain_model
    |---bytetrack_x_mot17.pth.tar
    |---halpe26_fast_res50_256x192.pth
    |---trained.pt
```

### Getting Started
```
python demo.py --config cfg_files/demo.yaml
```

## Evaluation on GigaCrowd dataset with Docker

### Step 1. Download dataset
Download the test set from [GigaVision challenge website](https://www.gigavision.cn/track/track?nav=GigaCrowd&type=nav) and put the images in <YOUR/PATH> with the following structure.

### Step 2. Load docker image
You can obtain the docker image from [here](https://pan.baidu.com/s/1Ye9VQ3vOx4qahSOdY81q4w?pwd=k77l).
```
tar -zxvf gigacrowd.tar.gz
```
```
docker load < gigacrowd.tar
```

### Step 3. Run the code
```
docker container run --gpus all --rm \
-it -v <YOUR/PATH>/images:/workspace/data/GigaCrowd/images \
boycehbz/gigacrowd:latest
```
Then you can execute the reconstruction. The SMPL parameters for each person will be saved in ```/workspace/output/final/gigacrowd/images```.
```
python main.py
```

When the fitting is finished, the ```predict.json``` can be generated in ```/workspace/output/results/results``` with the following command.
```
python generate_json.py
```

## Citation
If you find this code useful for your research, please consider citing the following paper.
```
@misc{huang2023crowdrec,
title={CrowdRec: 3D Crowd Reconstruction from Single Color Images}, 
author={Buzhen Huang and Jingyi Ju and Yangang Wang},
year={2023},
eprint={2310.06332},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@inproceedings{grouprec,
title={Reconstructing Groups of People with Hypergraph Relational Reasoning},
author={Huang, Buzhen and Ju, Jingyi and Li, Zhihao and Wang, Yangang},
booktitle={ICCV},
year={2023},
}
```

## Acknowledgments
Some of the code are based on the following works. We gratefully appreciate the impact it has on our work. 
[GroupRec](https://github.com/boycehbz/GroupRec) 
[CLIFF](https://github.com/huawei-noah/noah-research/tree/master/CLIFF) 
[ByteTrack](https://github.com/ifzhang/ByteTrack) 
[YOLOX](https://github.com/Megvii-BaseDetection/YOLOX)

CrowdRec

CrowdRec 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

CrowdRec

CrowdRec 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README