Adapted-Center-and-Scale-Prediction
所属分类：C#编程
开发工具：Python
文件大小：0KB
下载次数：0
上传日期：2021-10-30 17:20:18
上 传 者：
sh-1993
说明： Pytorch实现“自适应中心_尺度预测：更稳定_更准确”
(Pytorch implementation of "Adapted Center _and_ Scale Prediction: More stable _and_ More Accurate")
文件列表:
config.py (1594, 2021-10-30)
data/ (0, 2021-10-30)
data/citypersons/ (0, 2021-10-30)
data/citypersons/annotations/ (0, 2021-10-30)
data/citypersons/annotations/__init__.py (1, 2021-10-30)
data/citypersons/annotations/anno_train.mat (387145, 2021-10-30)
data/citypersons/annotations/anno_val.mat (77204, 2021-10-30)
data/citypersons/images/ (0, 2021-10-30)
dataloader/ (0, 2021-10-30)
dataloader/__init__.py (1, 2021-10-30)
dataloader/data_augment.py (5494, 2021-10-30)
dataloader/load_data.py (2314, 2021-10-30)
dataloader/loader.py (10294, 2021-10-30)
eval_city/ (0, 2021-10-30)
eval_city/__init__.py (1, 2021-10-30)
eval_city/cocoapi/ (0, 2021-10-30)
eval_city/cocoapi/LuaAPI/ (0, 2021-10-30)
eval_city/cocoapi/LuaAPI/CocoApi.lua (10726, 2021-10-30)
eval_city/cocoapi/LuaAPI/MaskApi.lua (10156, 2021-10-30)
eval_city/cocoapi/LuaAPI/cocoDemo.lua (791, 2021-10-30)
eval_city/cocoapi/LuaAPI/env.lua (436, 2021-10-30)
eval_city/cocoapi/LuaAPI/init.lua (498, 2021-10-30)
eval_city/cocoapi/LuaAPI/rocks/ (0, 2021-10-30)
eval_city/cocoapi/LuaAPI/rocks/coco-scm-1.rockspec (821, 2021-10-30)
eval_city/cocoapi/MatlabAPI/ (0, 2021-10-30)
eval_city/cocoapi/MatlabAPI/CocoApi.m (13117, 2021-10-30)
eval_city/cocoapi/MatlabAPI/CocoEval.m (22360, 2021-10-30)
eval_city/cocoapi/MatlabAPI/CocoUtils.m (16584, 2021-10-30)
eval_city/cocoapi/MatlabAPI/MaskApi.m (4979, 2021-10-30)
eval_city/cocoapi/MatlabAPI/cocoDemo.m (1204, 2021-10-30)
eval_city/cocoapi/MatlabAPI/evalDemo.m (1808, 2021-10-30)
eval_city/cocoapi/MatlabAPI/gason.m (2395, 2021-10-30)
eval_city/cocoapi/MatlabAPI/private/ (0, 2021-10-30)
eval_city/cocoapi/MatlabAPI/private/gasonMex.cpp (9227, 2021-10-30)
eval_city/cocoapi/MatlabAPI/private/gasonMex.mexa64 (38020, 2021-10-30)
eval_city/cocoapi/MatlabAPI/private/gasonMex.mexmaci64 (41452, 2021-10-30)
... ...
The codes are about my paper [**Adapted Center and Scale Prediction: More stable and More Accurate**](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/). This is the pytorch implementation. However, because my paper is based on [**High-level Semantic Feature Detection: A New Perspective for Pedestrian Detection**](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/) and this is my first try to begin a computer vision project, I choose to use and change the codes from [**CSP PyTorch Implementation**](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/). Many thanks for his contribution！
Welcome to [**my website**](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/)
# ACSP PyTorch Implementation
## Requirement
* Python 3.6
* Pytorch 0.4.1.post2
* Numpy 1.16.0
* OpenCV 3.4.2
* Torchvision 0.2.0
## Reproduction Environment
* Test our models: One GPU with about 4G memory.
* Train new models: Two GPUs with 32G memory per GPU.(If you do not have enough GPU memory resources, please resize the input to 640x1280, it yields slightly worse performance, though.)
## Installation
You can directly get the codes by:
```
git clone https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction.git
```
## Preparation
1. CityPersons Dataset
You should download the dataset from [here](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/https://www.cityscapes-dataset.com/downloads/). From that link, leftImg8bit_trainvaltest.zip (11GB) is used. We use the training set(2975 images) for training and the validation set(500 images) for test. The data should be stored in `./data/citypersons/images`. Annotations have already prepared for you. And the directory structure will be
```
*DATA_PATH
*images
*train
*aachen
*bochum
...
*val
*frankfurt
*lindau
*munster
*test
*berlin
*bielefeld
...
*annotations
*anno_train.mat
*anno_val.mat
...
```
2. Pretrained Models
The backbone of our ACSP is modified ResNet-101, i.e. replacing all BN layers with SN layers. You can download the pretrained model from [here](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/https://pan.baidu.com/s/1rK-ukAjEIPql2ECi38hRbQ). It is provided by the author of [Switchable Normalization](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/https://github.com/switchablenorms/Switchable-Normalization). The weights will be stored in `./models/`.
3. Our Trained Models
We provide two models:
[ACSP(Smooth L1)](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/https://pan.baidu.com/s/1p2IF7nI6dOhpmSvXLFsxlA)(code: ydc1): **Reasonable 10.0%; Heavy 46.1%; Partial 8.8%; Bare 6.7%**.
[ACSP(Vanilla L1)](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/https://pan.baidu.com/s/1zZP3brc1FvMrcmPo7Fx-Tg)(code: 4oa2): **Reasonable 9.3%; Heavy 46.3%; Partial 8.7%; Bare 5.6%**.
They should be stored in `./models/`.
4. Compile Libraries
Before running the codes, you must compile the libraries. The followings should be accomplished in terminal. If you are not sure about what it means, click [here](https://github.com/WangWenhao0716/Adapted-Center-and-Scale-Prediction/blob/master/https://linuxize.com/post/linux-cd-command/) may be helpful.
```
cd util
make all
```
## Training
`python train.py`
or
`CUDA_VISIBLE_DEVICES=x,x python train.py --gpu_id 0 1`
## Test
`python test.py`
## Citation
If you think our work is useful in your research, please consider citing:
```
@article{wang2020adapted,
title={Adapted Center and Scale Prediction: More Stable and More Accurate},
author={Wang, Wenhao},
journal={arXiv preprint arXiv:2002.09053},
year={2020}
}
```
## Note
This is my first computer vision project. If you have any questions or there are something wrong in my codes, feel free to contact me: wangwenhao@buaa.edu.cn.
近期下载者：
相关文件：
收藏者：