# Semantics Sharing
This repository is the offical release for the paper <a href="https://arxiv.org/pdf/1909.07038.pdf"><strong>Boosting Real-Time Driving Scene Parsing with Shared Semantics</strong></a>. The video introducing the paper is available <a href="https://youtu.be/Od1WVaqqt1o">here</a>.
In this paper, we proposed a deep neural network for real-time scene parsing which leverages shared semantics between cameras with different perspectives and overlapped views. Compared with traditional approaches, our method mainly boosts the scene parsing task for multi-camera systems in the following two aspects: 1) reduce the computation load for cameras with narrower field of views (FoVs), and 2) improve the scene parsing quality for cameras with broader FoVs.
<p align="center">
<img src="resource/motivation.jpg" width="600">
</p>
## Network Architecture
![](resource/network.jpg)
## Requirements
- Python 3.6
- PyTorch 1.2
- spatial-correlation-sampler
- kornia
## Dataset
- image pairs
The dataset consists of image pairs from the cameras with 60 degree horizontal FoV and 120 degree horizontal FoV. The format of the file names should following the example images in the [data/image_train](data/image_train) folder: **fov60_00001** & **fov120_00001**.
Note that the number of the same frame must be the same. If more images are added, please increase the index number in the file name.
- label
The category that is not included in the loss is labeled -1.
## Train
Run the train script with different models and fix some corresponding modules:
```
python3 train.py --model-mode [MODEL_NAME] \
--fixed_mobilenetv3 [True] --fixed_pwcnet [True] --fixed_FFM120 [False] --fixed_FFM60 [True]
```
The ``[MODEL_NAME]`` can be chosen from: mobilenetv3, mobilenetv3_loosely, mobilenetv3_tightly. More training details and parameters setting can be found in the paper.
## Evaluate
Run the eval script with different models and corresponding model path:
```
python3 train.py --model-mode [MODEL_NAME] --resume [MODEL_PATH] --save-pre-path [SAVING_PATH] --combined [False]
```
The ``[SAVING_PATH]``is the path to save the inference results of the validation set. The form of the results is determined by the parameter **--combined**. If it is *True*, the output will consist of four RGB images.
## Test
With the provided trained models, please follow the usages below to perform the testing:
- Prepare image data and model files
The example input images are provided in the [data/image_test](data/image_test) folder.
The trained models can be downloaded from the following links: [[mobilenetv3](https://drive.google.com/file/d/1j2oiqkg9MfzeaGTF_8AO1Trzle4Xpmhm/view?usp=sharing)], [[loosely-coupled](https://drive.google.com/file/d/15hulONRoEEUrMIJ0BCpG6b_VYxA1TRDI/view?usp=sharing)], [[tightly-coupled](https://drive.google.com/file/d/1U28ceVAHXq9p5wZyozZuc5cWIrFj1I4Q/view?usp=sharing)]. The models should be put in the [scripts/checkpoint](scripts/checkpoint) folder for testing.
- Inference
Run the test script with different models and corresponding PyTorch model files:
```
python3 test.py --model-mode [MODEL_NAME] --resume [MODEL_PATH]
```
- Comparisons
![](resource/results.jpg)
## License and Citation
All code and other materials (including but not limited to the paper, figures, and tables) are provided for research purposes only and without any warranty. Any commercial use requires our consent. When using any parts of the code package or the paper (<i>Boosting Real-Time Driving Scene Parsing with Shared Semantics</i>) in your work, please cite the following paper:
```
@misc{xiang2019boosting,
title={Boosting Real-Time Driving Scene Parsing with Shared Semantics},
author={Zhenzhen Xiang and Anbo Bao and Jie Li and Jianbo Su},
year={2019},
eprint={1909.07038},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
## References
- <a href="http://openaccess.thecvf.com/content_ICCV_2019/papers/Howard_Searching_for_MobileNetV3_ICCV_2019_paper.pdf">MobileNetV3</a>:
```
@InProceedings{Howard_2019_ICCV,
author = {Howard, Andrew and Sandler, Mark and Chu, Grace and Chen, Liang-Chieh and Chen, Bo and Tan, Mingxing and Wang, Weijun and Zhu, Yukun and Pang, Ruoming and Vasudevan, Vijay and Le, Quoc V. and Adam, Hartwig},
title = {Searching for MobileNetV3},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
year = {2019}
}
```
- <a href="http://openaccess.thecvf.com/content_cvpr_2018/papers/Sun_PWC-Net_CNNs_for_CVPR_2018_paper.pdf">PWC-Net</a>:
```
@InProceedings{Sun2018PWC-Net,
author = {Deqing Sun and Xiaodong Yang and Ming-Yu Liu and Jan Kautz},
title = {{PWC-Net}: {CNNs} for Optical Flow Using Pyramid, Warping, and Cost Volume},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2018}
}
```
## Acknowledgments
- [Tramac/Lightweight-Segmentation](https://github.com/Tramac/Lightweight-Segmentation): MobileNetV3 for semantic segmentation.
- [RanhaoKang/PWC-Net_pytorch](https://github.com/RanhaoKang/PWC-Net_pytorch): optical flow estimation.
- [ClementPinard/Pytorch-Correlation-extension](https://github.com/ClementPinard/Pytorch-Correlation-extension): PyTorch implementation of Corrleation Module.
- [nianticlabs/monodepth2](https://github.com/nianticlabs/monodepth2): unsupervised losses.
- [kornia/kornia](https://github.com/kornia/kornia): PyTorch implementation of homography warping.
## Contact
Zhenzhen Xiang ([email protected]), Anbo Bao ([email protected])
没有合适的资源?快使用搜索试试~ 我知道了~
这是“使用共享语义提升实时驾驶场景解析”的官方存储库.zip
共80个文件
py:44个
jpg:13个
png:12个
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 148 浏览量
2023-03-31
23:10:34
上传
评论
收藏 10.3MB ZIP 举报
温馨提示
这是“使用共享语义提升实时驾驶场景解析”的官方存储库
资源推荐
资源详情
资源评论
收起资源包目录
这是“使用共享语义提升实时驾驶场景解析”的官方存储库.zip (80个子文件)
SemanticsSharing-master
data
image_train
fov60_02450.jpg 855KB
fov60_21000.jpg 509KB
fov120_21000.jpg 459KB
fov120_02450.jpg 679KB
test_results
mobilenetv3_large_pwcflow_yuanqu
fov120_00001_mobilenetv3.png 14KB
fov120_00001_mobilenetv3_loosely.png 14KB
fov60_00001_mobilenetv3_loosely.png 15KB
fov60_00001_mobilenetv3.png 14KB
fov60_00001_mobilenetv3_tightly.png 14KB
fov120_00001_mobilenetv3_tightly.png 14KB
image_label
fov60_02450.png 13KB
fov60_21000.png 9KB
fov120_02450.png 13KB
fov120_21000.png 10KB
image_val
fov60_02450.jpg 855KB
fov60_21000.jpg 509KB
fov120_21000.jpg 459KB
fov120_02450.jpg 679KB
image_test
fov60_00001.jpg 454KB
fov120_00001.jpg 410KB
resource
network.jpg 208KB
results.jpg 474KB
motivation.jpg 292KB
requirements.txt 63B
light
__init__.py 0B
nn
__init__.py 40B
loss.py 4KB
basic.py 4KB
data
__init__.py 298B
yuanqu.py 12KB
map
flow_vis.py 4KB
pwcnet_model_mobilenetv3.py 13KB
pwcnet_args_mobilenetv3.py 793B
pwcnet_flow_utils.py 4KB
pwcnet_args.py 628B
pwcnet_modules.py 7KB
pwcnet_model.py 10KB
correlation.py 15KB
pwcnet_modules_mobilenetv3.py 9KB
warp.py 3KB
warp_xy.py 3KB
utils
__init__.py 0B
distributed.py 9KB
load_model.py 3KB
metric.py 3KB
logger.py 1KB
lr_scheduler.py 7KB
visualize.py 5KB
model
base_model
__init__.py 28B
mobilenetv3.py 6KB
debug.py 6KB
__init__.py 367B
mobilenetv3_pwcnet_seg.py 16KB
base.py 2KB
.gitignore 2KB
pre_train
optical_flow
utils.py 357B
predict.sh 136B
losses.py 6KB
main.py 19KB
LICENSE 1KB
dataset.py 11KB
model.py 5KB
summary.py 3KB
losses_unsupervised.py 6KB
modules.py 6KB
logger.py 2KB
flow_utils.py 4KB
example
flow_warped
CAM-120_homo_warped_flow_warped.png 488KB
pred.txt 40B
CAM-60_GT.jpeg 183KB
CAM-120_homo_warped.jpeg 150KB
flow_vis
CAM-120_homo_warped.png 111KB
flow
CAM-120_homo_warped.flo 3.07MB
README.md 648B
warp.py 3KB
README.md 6KB
scripts
checkpoint
README.md 64B
eval.py 13KB
train.py 19KB
test.py 7KB
共 80 条
- 1
资源评论
快撑死的鱼
- 粉丝: 1w+
- 资源: 9154
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功