InnoTIS

InnoTIS 是 Innodisk and Aetina 用來提供 Aetina Server 運行AI模型的效果，我們結合了 NVIDIA Triton Inference Server 的技術讓使用者可以透過gRPC的方式傳送資料到我們的 Aetina Server 進行 AI 推論進而取得辨識結果。

---

Feature

• Custom version which only have three model could use.
  1. DENSENET_ONNX ( NVIDIA sample )
  2. YOLOV4 ( COCO Dataset )
  3. YOLOV4_WILL ( Can Detect Does People Wear A Mask )
• To modify your custom code (*.cpp *.h), please visit my notion。
• I use gRPC in innotis-server, HTTP service will open but disable to use.

---

How to use？

1. Install NVIDIA Driver and Docker

   1. Build Triton on Windows ( With Docker )
   2. Build Triton on Ubuntu ( With Docker )

2. Run innotis-server

   1. Download innotis-server

      $ git clone https://github.com/MaxChangInnodisk/innotis-server.git
      $ cd innotis-server

   2. Run init.sh ( Only need first time )

      $ ./init.sh

      • Please remind the ip of server, you can alos find it in server_ip.txt.
      • The folder (build/, triton-deploy/) will be generated
   3. Run run.sh

      $ ./run.sh

      • Make sure GRPC and HTTP service is started.

3. Run innotis-client ( with another terminal )

   Github: innotis-client

   • DockerHub: pull image & run container from docker hub

     $ docker run --rm -p 5000:5000 -t maxchanginnodisk/innotis

   • Dockerfile: you can also build from docker file
     • Please visit innotis-client to get more information.
   • Miniconda: virtual environment might be a great idea for developer
     • Please visit innotis-client to get more information.

4. Open browser and enter url ( localhost:5000 ).

   • Triton IP must be modify to <server_ip>, you can find <server_ip> in "server_ip.txt" which will be generated when run init.sh

5. Have fun.

   

---

Reference

Thanks for:

• YOLOv4 on Triton Inference Server with TensorRT
• TensorRTx