potato_proj🥔

A ML project detect the 💉potato early blight and late blight disease.
🤖

Dataset：
https://www.kaggle.com/datasets/arjuntejaswi/plant-village?resource=download

system: windows10 (with cuda11.3)

problems & solutions

• python version higher than python3.9, can't find tensorflow2.5.
  (some version of tensorflow didn't have tensorflow.keras.layers.experimental.)

• when install pakages, notice the compatibility of dependencies and packages.
  Use commands like pip check.

• Encounter cuda problem, when training model.
  Get hint from notebook Browser userInterface: Kernel Restarting The kernel for training/training.ipynb appears to have died.It will restart automatically. if without any hints else, try to check the notebook running logs, in CLI, find massage like: Could not locate zlibwapi.dll. Please make sure it is in your library path so we try to find zlibwapi.dll. From NVIDIA official website below, I get some clues.
  https://forums.developer.nvidia.com/t/zlib-dll-for-latest-cudnn-in-official-install-guide-is-missing/197630/6
  If you can't find file zlibwapi.dll from where they find zlibwapi.dll file directories. Try to search all the C desk ,u will find a lot of zlibwapi.dll file. copy any one of them paste it to your directory like C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\bin. then, restart your kernel try to train again.

• Working on FastAPI error below occur.
  ImportError: cannot import name 'ParamSpec' from 'typing_extensions'.
  Try pip uninstall fastapiand pip install --no-cache fastapi.

• Docker problems maybe due to the ports num, typos like me 8501:8501-->8501:8051
  😫😫😫

Start Step & Structure

(Here use Jupyter with python -m notebook in folder /training)
tf dataset
Data augmentation(1.keras.layers.experimental.preprocessing & 2.ImageDataGenerator api)
↓
CNN (cache|prefetch)
↓
(from here I use pycharm in folder /api) Deployed via tf-serving in Docker to FastAPI.
run main-tf-serving.py,
open Docker run with:

docker run --rm -it -p 8502:8502 -v D:/ml_workspace/potato_proj:/potato_proj tensorflow/serving --rest_api_port=8502 --model_config_file=/potato_proj/models.config

You can test with Postman.
↓
(Here I use VSCode for Frontend and PyCharm for Backend) React JS (Build UI and interact with)
Node.js (JavaScript running env) run Node.js in folder /frontend with:

npm run start

Then run the /api/main.py in VSCode ↓
GCP
↓
React Native (mobile application, get stuck here, will do it afterward)