sarthakpranesh / plantrecog

Free Plant Recognition App + Public API + Models

PlantRecog aims to become or lead to the goto "Free Plant Recognition" solution. Why? As of today, 30th August 2021, there aren't many online free services or apps available to solve this problem. The existing solutions are mostly paid and have high subscription amounts per year. I want to create a fairly basic but usable set of services (app + api + models) for people who want to develop tools that recognize plants.

The current version of the project is able to recognize 299 plants using their flowers. This flower recognition model and API developed under this project are available freely in the repository. They can be self hosted by anyone who wants to use these services for building their own apps.

App

PlantRecog is now available on Google Play Store

Or get the latest release from GitHub Release Page

Recognition Service

This is the documentation for the server API that is used in PlantRecog. This server can be hosted online to develop apps that require such a service.

Request	Payload	Description
get "/"	{"message":"Server Up and Fine"}	Index route to make sure server is running, Heroku puts the server to sleep so it'll be great to call this in start of your app to wake the server up
get "/details"	{"message":"Success","payload":{"recognized":["Abutilon","Acacia",...,"Zenobia","Zinnia"]}}	Get all the recognized classes for the latest model available on the server
post "/predict" content-type="multipart/form-data" parameter="image"	{"messages":"Success","payload":{"predictions":[{"name":"MorningGlory","score":0.38581109046936035},{"name":"Acacia","score":0.14158271253108978},{"name":"MoonflowerVine","score":0.12431787699460983},{"name":"LilyoftheValley","score":0.06644751876592636},{"name":"FrangipaniFlower","score":0.062477629631757736}]}}	Post plant image using multipart/form-data, parameter name should be image and the route will provide the top 5 prediction for the plant image

Other services used

Gyan API - https://github.com/sarthakpranesh/Gyan

Issues

If you come across any issues or have feature requests please open them here

Made with ♥

plantrecog's Issues

Generate a better model v0.2.0

Goals

• Longer training time
• Better accuracy

Finish UI/UX of App

Reliability ?

I think this app lacks a bit of precision 😅
Is this normal? Is there any way to contribute to increase the accuracy?

Find and polish minor details

Open to all suggestions.
Install the app -> Try it out -> Suggest improvements (down in comments) -> creat pull requests -> :>

Interested in getting this up and running again

Any pointers?

PlantRecog v2: Things to do for next major release

Todo:

• Move ML model prediction from external API to inside app itself
• Look for improved database for training the model
• Upgrade expo to latest version

Add more...

Add Firebase Analytics to track app usage

cant docker build on m1 mac

cant docker build on m1 mac, can you help?
Error: Sorry! Volta currently only provides pre-built binaries for x86_64 architectures.
Error: The Node.js native addon module (tfjs_binding.node) can not be found at path: /Users/matvejivanejcik/Documents/projects/PlantRecog/server/node_modules/@tensorflow/tfjs-node/lib/napi-v8/tfjs_binding.node.
Please run command 'npm rebuild @tensorflow/tfjs-node --build-addon-from-source' to rebuild the native addon module.

Restructure APIs for lower request count per prediction v1

Routes on the server can be shrunk/merged together for lowering the request count needed from app start to making the first prediction.

Document v1 API

Document all the routes in v1 of the API

Finalize API endpoints

Initial thoughts on all required APIs for v1, will update the below list if I feel some other endpoints are required

• route to get details (version, classes recognized) about the model using the version number
• predict route (already present but look for improvements)
• route to scrape google and retrieving images using plant class ( can scrape images of the predicted flower to display other images to the user )
• route to scrape Wikipedia for plant details using plant class
• refactor all routes

Moving towards all in App computation

What
Move towards doing all processing, whether it be ML computations or scrapping the web, inside app itself.

Why
Managing your own server or using services like AWS is expensive and not very feasible for projects like this one.

Tasks

• Moving ML processing from server side API into App
• Scrapping web in app instead of using GyanApi

Camera buttons get cut on devices with notch

Devices that have a notch lead to a black header in place of the statusbar and the screen starts below the notch. This leads to buttons shifting below the bottom sheet, and getting covered partially.

Add common type declaration for app

Get simiar images route provide same image

The server route "/images/:name" is supposed to provide similar but different flower images, but right now it just provides one image duplicated in the array.

Update the API documentation in the readme

After #10 the documentation of the server API present in the readme needs to be updated.