# Semantic TSDF
Semantic TSDF for self-driving static scene reconstruction.
TSDF (truncated signed distance function) can be used to reconstruct the static scene around a self-driving car. Since it is fast and optimizable by GPU, we prefer to use TSDF rather than fused point cloud to represent the world. However, we must filter out dynamic objects that may hinder our reconstruction. Since motion-based methods will still capture cars that stop for a while at the red light, we would like to use semantic segmentation networks to filter out the objects that are possibly dynamic.
Below is the reconstruction result on KITTI sequence 0106. Dynamic objects such as cars (including parked cars), bikes, and pedestrians are successfully filtered out.
![KITTI Sequence 0106](examples/0106_color.gif)
![Reconstruction Example](examples/0106_tsdf.gif)
Please refer to the LICENSE file for inherited Licenses.
## Dependencies
pytorch 1.0+
Optional: GPU acceleration requires an NVIDA GPU with [CUDA](https://developer.nvidia.com/cuda-downloads) and [PyCUDA](https://developer.nvidia.com/pycuda)
```shell
pip install pycuda
```
## Data generation
This step is completed by default. To generate a new data sequence, create a `KITTI` directory and unzip the downloaded raw KITTI data. The directory will look like this
```text
- KITTI
| - 2011_09_26
| - 2011_09_26_drive_XXXX_extract
| - 2011_09_26_drive_XXXX_sync
| - calib_cam_to_cam.txt
| - calib_imu_to_velo.txt
| - calib_velo_to_cam.txt
```
Then, simply run
```shell
python data_parser.py
```
Note: depth images are saved as 16-bit PNG in millimeters.
Note: We currently only provide pretrained model on cityscape. For KITTI, you may need to fine-tune yourself. The `kitti.pkl` file provided is only a dummy file.
## Demo
```shell
python demo.py
```
The result is save as a `.ply` file, and we recommend to use `Meshlab` to render it with colors.
没有合适的资源?快使用搜索试试~ 我知道了~
资源推荐
资源详情
资源评论
收起资源包目录
场景重建_用于自动驾驶静态场景三维重建的语义TSDF算法_GPU优化_优质项目实战.zip (16个子文件)
场景重建_用于自动驾驶静态场景三维重建的语义TSDF算法_GPU优化_优质项目实战
tsdf
utils.py 2KB
__init__.py 0B
tsdf.py 14KB
data_parser.py 6KB
examples
0106_tsdf.gif 5.03MB
0106_color.gif 17.8MB
network
__init__.py 0B
modules
__init__.py 110B
rfblock.py 7KB
bn.py 3KB
misc.py 4KB
model.py 8KB
checkpoints
cityscape.pkl 66.11MB
kitti.pkl 0B
demo.py 3KB
README.md 2KB
共 16 条
- 1
资源评论
极智视界
- 粉丝: 2w+
- 资源: 1459
下载权益
C知道特权
VIP文章
课程特权
开通VIP
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
最新资源
- 本科毕业设计基于C# wpf人脸识别的考勤系统的设计与实现源码.zip
- 基于Ruoyi+uniapp实现学生考勤系统 学生考勤源码+项目说明.zip
- feae6bc968ca68a099455d8b8a8dea35
- 基于Pytorch训练CIRAR10上分类算法.zip
- Pytorch-pytorch深度学习教程之Tensorboard.zip
- 基于C++和Python开发yolov8-face作为人脸检测器dlib作为人脸识别器的人脸考勤系统源码+项目说明.zip
- Pytorch-pytorch深度学习教程之变分自动编码器.zip
- Pytorch-pytorch深度学习教程之神经风格迁移.zip
- Pytorch-pytorch深度学习教程之深度残差网络.zip
- Pytorch-pytorch深度学习教程之循环神经网络.zip
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
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