DEVNET Project 2: Health Detection via Image Classification

Installation

Installation is very simple, just install the needed dependencies.

Use the following index-urls for the desired torch backend:

• cpu: https://download.pytorch.org/whl/cpu
• cuda: https://download.pytorch.org/whl/cu124
• rocm: https://download.pytorch.org/whl/rocm6.2

python3 -m pip install torch==2.5.0 torchvision==0.20.0 torchaudio==2.5.0 --index-url https://download.pytorch.org/whl/cpu
python3 -m pip install -r requirements.txt

Training

While the included model.pth weights file should be enough for most, you can train the model from scratch or further with the train.py and pretrain.py file. The script will download the kagglehub dataset and train based on those images.

To start, we need to preprocess our images:

python3 pretrain.py

Now we can train the model:

python3 train.py

Training Options:

The training script will automatically use the best device available, and you can tweak the epoch and batch size.

$ python train.py -h
usage: Trainer [-h] [-e EPOCHS] [-b BATCH] [-s]

Trains the model based on preprocessed images

options:
  -h, --help            show this help message and exit
  -e EPOCHS, --epochs EPOCHS
                        The number of epochs to train on
  -b BATCH, --batch BATCH
                        The size of the batches
  -s, --silent          Disables all print statements

After training there will be a new or modified file model.pth which contains the weights for the model.

Running The Model

Via Premade File

The script test.py is made to easily test the model on supplied images:

$ python test.py -h
usage: test.py [-h] [-p PATH] [-w WEIGHTS] [--debug]

options:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  The path to the image or image directory
  -w WEIGHTS, --weights WEIGHTS
                        The path to the weights of the model
  --debug               Whether to enable very verbose and helpful debugging

Via API

There is also an api in the HealthModel.py file which allow you to access an interposer and evaluate batches or urls to images.

import cv2
from HealthModel import HealthModel

model = HealthModel("model.pth")

image = cv2.imread("PATH TO IMAGE FILE")
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
pred = model.predict(image, multi_leaf=..., _debug=...)

print(f"The image is {'Unhealthy' if pred[0] else 'Healthy'} ({pred[1] * 100:.2f}%)")

The .predict() method will return the health of the plant as an int (0 or 1 (0 is healthy)) and a confidence value (percent from 0.0 to 1.0).

Web UI

python3 webserver.py

Then open localhost