Transformer是一种基于注意力机制，完全不需要递归或卷积网络的序列预测模型，且更易于训练
介绍了Gated-RNN/LSTM的基本逻辑[[Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling]]，指出:
这种固有的顺序性质阻碍了训练示例中的并行化，这在较长的序列长度上变得至关重要，因为内存限制限制了示例之间的批处理，虽然后续有相关工作优化了一些性能，但是基本的限制并没有解除。
https://github.com/hkproj/pytorch-transformer/
https://www.youtube.com/watch?v=ISNdQcPhsts
关于Transformer 中的LayerNorm: https://dkleine.substack.com/p/understanding-layer-normalization?utm_campaign=post&utm_medium=web
参考: https://forum-zh.obsidian.md/t/topic/292
总结了各文章之间的 关联度，太直观了
Citexs Paperpicky
领域文献调研分析网络图将推荐文献可视化，在这里可以很容易看到文献之间的引用和被引用的关系。每个圆球代表一篇文章，圆球大小与共引次数正相关，从AI推荐文献、经典文献、核心文献三个维度给大家推荐出关联文献。
（1）AI推荐文献是根据所输入的关键词，智能筛选出与关键词高度关联的文献，在这里大家可以看到近期发表的研究成果；
（2）经典文献是基于推荐文献共引用关系筛选出的文献，一般都是该领域的必读文献，也就是说只要你研究该领域，这些文献肯定要读；
（3）核心文献则是AI推荐文献和经典文献重叠共引的部分。
参考仓库: https://github.com/rzeldent/esp32cam-rtsp
Stream video through wifi using ESP-32-CAM.
The video sources can be accessed by an ip address.
Install PlatformIO plugin in vscode.
Clone the repository
1 | git clone --recursive https://github.com/rzeldent/esp32cam-rtsp.git |
Use vscode pio to open the project. Wait pio till its configuration is done.
Change the default_envs settings, here I use esp32cam_ai_thinker
If no
default_envsis specified, pio will build the project for all platforms
Here we can build and upload the program to the board.
Connect to ESP**** WiFi, and visit http://192.168.4.1 to configure the wifi settings of the board. I choose to my phone’s hot spot.
Then open the monitor to check the ip address (you need to connect your computer to the same LAN (local area network) to visit the ip)
Visit the 192.168.23.142 and you will see the page (similar with the page of http://192.168.4.1):
Click rtsp://192.168.23.197:554/mjpeg/1 and you will see the streaming video:
生无可恋做横向.jpg
Guide: https://docs.espressif.com/projects/esp-idf/en/stable/esp32/get-started/linux-macos-setup.html#
I directly download the archive with all the submodules included: https://github.com/espressif/esp-idf/releases/tag/v5.3.1
This archive can also be downloaded from Espressif’s download server: https://dl.espressif.com/github_assets/espressif/esp-idf/releases/download/v5.3.1/esp-idf-v5.3.1.zip
cd into the unzip folder (The installation will fail in conda venv)
1 | conda deactivate |
In ~/.zshrc
1 | alias get_idf='. $HOME/esp/esp-idf-v5.3.1/export.sh' |
Then each time I need to setup esp32 development environment, I only need to type get_idf.
Some useful commands
1 | idf.py set-target esp32 |
事实证明pio是最方便的。。
事前安装过pio的vscode插件，直接打开pio的esp32项目就直接可以编译上传以及查看串口监视器。
希望可以每天看两个小时论文
爱心屋签到： aixinwu.sjtu.edu.cn/products/asw-store
中午12点抢羽毛球场
午/晚饭后乌灵💊
每日二GRISSO💊
| Time | Monday | Tuesday | Wednesday | Thursday | Friday | Saturday | Sunday |
|---|---|---|---|---|---|---|---|
| 08:00 | 周末睡懒觉～ | ||||||
| 09:00 | 读论文 | 读论文 | 读论文 | Arch Update | |||
| 10:00 | 组会 | Seminar | |||||
| 11:00 | |||||||
| 12:00 | Friday afternoon | ||||||
| 13:00 | |||||||
| 14:00 | ME6801(DXY402) | 新中特(CRQ214) | E-Guitar | ||||
| 15:00 | |||||||
| 16:00 | ECE6903(DZY3-105) | ME6801(DXY402) | ECE6903(DZY3-105) | ||||
| 17:00 | 博客网维护 | ||||||
| 18:00 | VG501(DXY112) | Nvim Update | |||||
| 19:00 | E-Guitar | E-Guitar | E-Guitar | E-Guitar | E-Guitar | ||
| 20:00 | Movie | ||||||
| 21:00 | |||||||
| 22:00 | Reading | Reading | Reading | Reading | Reading | ||
| Credits: 3+3+2+1 |
Moved to [[2025 Winter&Spring Schedule]]
Moved to [[2025 Winter&Spring Schedule]]
Moved to [[2025 Winter&Spring Schedule]]
~/Book [[《学做工：工人阶级子弟为何继承父业》读书会p11]] (只看了一半) 🔒Done! 2024-11-27 🕸️ 2024 Fall > Routine > Non-routine > No Due
一款线上还在更新中的shader教学，有可交互代码，爆赞
https://thebookofshaders.com/
一些dalao写的shader：
https://www.shadertoy.com/
Shader Tutorials by Ronja
https://www.ronja-tutorials.com/
https://github.com/ronja-tutorials/ShaderTutorials?tab=readme-ov-file
Shader Tutorial For Beginner
https://github.com/Xibanya/ShaderTutorials
因为oh-my-live2d官网崩了，所以需要自己私有托管一下live2d服务才能让博客网上的小埋复活。(bonus:可以借此尝试自己制作一个live2d形象)
更于9/25，拖着很久没搞，然后官网自己复活了，好好好，直接快进到customized live2d。
