Stanford CS330: Class Project.

TLDChoiceNet: Quantitatively Choosing a Transfer Learning Dataset

We used a few datasets from Kaggle:

• New Plant Diseases Dataset: 256 x 256 RGB JPG images of healthy and unhealthy crop leaves
  • Replaced with TensorFlow Datasets plant_village dataset
• Plant Leaves for Image Classification: 6000 x 4000 RGB JPG images of healthy and unhealthy leaves from 12 plants
• BIRDS 450 SPECIES- IMAGE CLASSIFICATION: 224 x 224 RGB JPG images of bird species

Here's how to easily download them all with the Kaggle API:

kaggle datasets download -p data/plant-diseases --unzip vipoooool/new-plant-diseases-dataset
kaggle datasets download -p data/plant-leaves --unzip csafrit2/plant-leaves-for-image-classification
kaggle datasets download -p data/bird-species --unzip gpiosenka/100-bird-species

This project was developed using Python 3.10.

Here is how to create a virtual environment to work with this repo:

python -m venv venv
source venv/bin/activate
python -m pip install -r requirements.txt

We love quality code! If you do too, run these commands after creating the environment:

python -m pip install -r requirements-qa.txt
pre-commit install

Here is how you kick off tensorboard:

tensorboard --logdir training

Afterwards, go to its URL: http://localhost:6006/.