Naming is still in the works 😀
Add a nomination here: (Cherie, Ripe)
This project aims to leverage ML in determining the quality score of freshly picked coffee cherries being brought in to the wet mills by farmers/collection-agents in Africa.
The quality score would be leveraged to implement differential pricing of coffee cherries (high quality/ripeness scores imply higher purchasing prices).
Read more about the product here
This is a monorepo containing several codebases
| Codebase | Description |
|---|---|
| cherry_serve | React Native App |
| ripe | Pix2Pix GAN V2 |
| pix2pix | Pix2Pix GAN for Image segmentation |
| unet | Unet GAN for Image segmentation |
| serialized model | TfLite Model (latest) |
- main -> don't touch, create a branch, work on your feature, and submit a PR
First of all, this project is currently in very early stages of development, therefore these instructions may not be up to date.
For new React native developers
Install NodeJS, npm & yarn (nvm is a preferred way)
You can also follow the official React Native guide to getting setup: HERE
This project uses the managed expo workflow (for now), so you'll need to install the expo-cli
yarn global add expo-cli
Clone the project
git clone https://link-to-projectGo to the project directory
cd my-projectInstall dependencies
We're using Yarn for this project, do not use npm for the following commands
cd cherry_serve
yarn
You should now be all set to go, go ahead and run the dev server
yarn startYou now have a metro bundler running, you can start the app on Android or iOS
If you're already familiar with JavaScript, React and React Native, then you'll be able to get moving quickly! If not, we highly recommend you to gain some basic knowledge first, then come back here when you're done.
There have been 4 different methods applied in the experiments to perform semantic image segmentation on coffee cherry images.
- Plotting pixel colors in a 3d plot of HSV, RGB colorspaces to see if the images cluster automatically Here
- Using U-Net Here
- Using pix2pix GAN (tensorflow 1.4.1) Here
Current implementation in repo may be out of date and has no saved weights. Check Google drive link
(Might require special access) - Using a second pix2pix GAN (tensorflow + Keras) [Latest]
There are currently no tests
// TODO: Write testshttps://github.com/tensorflow/tfjs/blob/master/tfjs-converter/README.md