欢迎来到天眸(TianMouC)的项目主页
其他资源
借鉴人类视觉中双通路感知的基本原理，清华大学类脑计算研究中心发展了新一代异构互补融合的双通路类脑视觉感知芯片架构：天眸芯 (Tianmouc) 。
Tianmouc是世界首款多通路类脑视觉传感器，又称互补视觉传感器(CVS)，使用单个芯片即可灵活实现高速（10,000fps）、高动态（130dB）、高灵敏度（72%@530nm，NIR）、高精度（10bit）与低功耗、低带宽（相较于传统高速相机，具有90%带宽下降）的视觉感知，综合指标（FoM）位列世界一流。
DEMO视频示例
下载资料
1. 算法代码
Nature论文中使用的代码
可以从github中获取
git clone https://github.com/Tianmouc/tianmouc_paper_code.git
Python视觉工具包TianmoucV
公开版本为0.3.2，开发版本测试中
conda create -n tianmoucv --python=3.10
conda activate tianmoucv
conda install pytorch torchvision torchaudio pytorch-cuda=12.0 -c pytorch -c nvidia
pip install tianmoucv -i https://pypi.tuna.tsinghua.edu.cn/simple
TianmoucV文档见 TianMouCV算法库，安装中遇到的问题以及可能需要的资料请查阅 常见问题(Q&A)
2. 数据集
usb模组测试数据
你可以使用 tianmoucv/exmaple/data/test_data_read.ipynb 尝试这段数据
更多的数据集将在将来发布
Nature论文中使用的部分数据
你可以从 Zendo 中下载我们的数据，也可以在这里直接下载
解压数据后，使用软连接将数据连接到代码中的data文件夹
cd [N_pub_code]
ln -s [your dataset path] data
3. 图形用户界面
预编译GUI(USB驱动版本，tmdat数据)
USB版本的驱动预编译如下，支持ubuntu20/win10以上的系统使用，可以用作播放器
Version |
Download Link |
update Date |
|---|---|---|
ubuntu 20.04 x86 |
20240617 |
|
ubuntu 20.04 arm64 |
20240627 |
|
ubuntu 20.04 arm64 with CUDA |
20240403 |
|
ubuntu windows10 x86 |
20240319 |
预编译GUI(PCIE驱动版本,论文中使用的bin文件数据)
PCIE版本的驱动预编译如下，支持ubuntu20以上的系统使用，可以用作播放器
Version |
Download Link |
update Date |
|---|---|---|
ubuntu 20.04 x86 |
20240301 |
|
ubuntu 20.04 arm64 |
20240301 |
|
ubuntu 20.04 arm64 with CUDA |
20240301 |
引用和出版物
如果您在自己的工作中用到了天眸视觉算法库(Tianmoucv)，您可以按照下列格式进行引用：
@misc{Tianmouc}
其中的 YYYY-MM-DD 需要更改为您的工作使用的天眸视觉算法库(Tianmoucv)版本对应的最后一次代码修改日期。
文档索引
开源软件项目作者信息
Yihan Lin 林逸晗 For GUI/SDK/ISP-Algorithms/Data/Maintainace
Taoyi Wang 王韬毅 For GUI/SDK/Chip Design/Data/ISP-Algorithms
Welcome to TianMouC’s project homepage!
Other Resources
Based on the basic principle of dual-pathway perception in human vision, CBICR at Tsinghua University has developed a new generation of heterogeneous complementary dual-pathway vision chip architecture: Tianmouc.
Tianmouc is the world’s first multi-pathway brain-inspired vision sensor, also known as Complementary Vision Sensor (CVS), which realizes high speed (10,000fps), high dynamics (130dB), high sensitivity (72%@530nm, NIR), and high accuracy (10 bit) with low power consumption and low bandwidth (90% bandwidth degradation compared to traditional high-speed cameras) for visual perception, and the comprehensive perfoermance (FoM) is SOTA in the world.
DEMO video
Download contents
1. Algorithms Codes
Code used in the main paper
available in github repo
git clone https://github.com/Tianmouc/tianmouc_paper_code.git
Python Visualization Toolkit: TianmoucV
Public version is 0.3.2, development version in testing
conda create -n tianmoucv --python=3.10
conda activate tianmoucv
conda install pytorch torchvision torchaudio pytorch-cuda=12.0 -c pytorch -c nvidia
pip install tianmoucv -i https://pypi.tuna.tsinghua.edu.cn/simple
Documents can be found at TianMouCV算法库，Problems encountered during installation and information that may be required can be found at 常见问题(Q&A)
2. Dataset
USB module test data
You can try this data using tianmoucv/exmaple/data/test_data_read.ipynb
More datasets will be released in the future
Some of the data used in the Nature paper
You can found the demo data in Zendo or directly download here:
After unzipping the data, use a soft link to connect the data to the data folder in the code
cd [N_pub_code]
ln -s [your dataset path] data
3. Graphical user interface (GUI)
Pre-compiled GUI (USB driver version, ‘.tmdat’ format)
The USB version of the driver is pre-compiled as follows, support ubuntu20/win10 or above system use, can be used as a player
Version |
Download Link |
update Date |
|---|---|---|
ubuntu 20.04 x86 |
20240617 |
|
ubuntu 20.04 arm64 |
20240627 |
|
ubuntu 20.04 arm64 with CUDA |
20240403 |
|
ubuntu windows10 x86 |
20240319 |
Pre-compiled GUI (PCIE driver version, ‘.bin’ file data used in the Nature paper)
The PCIE version of the driver is pre-compiled as follows, supports ubuntu20+ system use, and can be used as a player
Version |
Download Link |
update Date |
|---|---|---|
ubuntu 20.04 x86 |
20240301 |
|
ubuntu 20.04 arm64 |
20240301 |
|
ubuntu 20.04 arm64 with CUDA |
20240301 |
Citations and Publications
If you have used the Tianmoucv Visual Algorithm Library (Tianmoucv) in your own work, you may cite it in the following format:
@misc{Tianmouc}
where YYYY-MM-DD needs to be changed to the date of the last code modification corresponding to the version of Tianmoucv used in your work.