Skip to content

HACC2018/ohia.ai

Repository files navigation

An open Hawaiian indigenous plant identification and aggregation application. Developed for the 2018 Hawaii Annual Code Challenge. Learn more about the challenge and our app at devpost.

Table of Contents

Overview

To get started, first read the introductory text under Quick Start Application. Or read the Data Collection, Machine Learning, and Results sections of this README.

When ready for more information about each step of the process, visit any of the following:

Quick Start Application

Our app was built for the iPhone 6 running on iOS version 12. However, deploying the app to an iPhone device is a long process, especially if your Mac is not up-to-date with the latest OS or Xcode version. Therefore, we recommend these quick start steps for getting up-and-running quickly on the browser. Note, however, that the main feature of the app, plant identification, will not work because the native iOS camera cannot run on the browser.

If you would rather follow the complete instructions and deploy to an iOS device, follow the steps under Complete Application Setup.

Getting Started

  1. Clone this repository by running the following command in a terminal: git clone [email protected]:HACC2018/ohia.ai.git.
  2. Change into the ohia.ai project directory: cd ohia.ai.
  3. When running anything related to the client or server, we require Python 2.7 to be installed, since that is the version that the tfjs-node package supports. Switch your terminal to use Python 2.7 (more information here).
  4. Switch your terminal to use Node 10.12.0 (more information here).

Install server dependencies and run the server

  1. From the project root directory, run npm install. Among other items, you should see "Downloading libtensorflow" and "Building TensorFlow Node.js bindings" run without error in your output.
  2. Create a .env file in the project root by copying the template in .env.sample.
  3. Source your environment by running: source .env.
  4. You should now see these variables in your environment by running env.
  5. Add the machine learning model weight files from our Google Drive by first downloading and unzipping weights.tar.gz using the command tar xzf weights.tar.gz, followed by replacing the /server/models/mobilenetv1-1.00 folder with the respective unzipped folder.
  6. Run the server: npm run dev-server. You should see the following output:
    ▶ npm run dev-server
    
    > [email protected] dev-server /path/to/ohia.ai
    > nodemon server/index.js
    
    [nodemon] 1.18.4
    [nodemon] to restart at any time, enter `rs`
    [nodemon] watching: *.*
    [nodemon] starting `node server/index.js`
    2018-11-07 07:25:31.386112: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.2 AVX AVX2 FMA
    Server for ohia.ai listening on port 3000
    

Install Vue.js and Quasar and run the client in a browser

  1. Open another terminal window and switch to Python 2.7 and Node 10.12.0.
  2. Source the environment: source .env.
  3. Install Vue CLI: npm install -g @vue/cli.
  4. Install Vue CLI addon: npm install -g @vue/cli-init.
  5. Install Quasar CLI: npm install -g quasar-cli.
  6. From the project root directory, install client dependencies: npm run install:client.
  7. Run npm run dev to view the app in a browser. You should see the following output:
    ▶ npm run install:client
    
    > [email protected] install:client ~/ohia.ai
    > cd client && npm install
    
    
    > [email protected] install ~/ohia.ai/client/node_modules/fsevents
    > node install
    
    [fsevents] Success: "~/ohia.ai/client/node_modules/fsevents/lib/binding/Release/node-v64-darwin-x64/fse.node" already installed
    Pass --update-binary to reinstall or --build-from-source to recompile
    added 1333 packages from 612 contributors and audited 16152 packages in 23.649s
    found 0 vulnerabilities
    
    (python2)
    
  8. Visit http://localhost:8080 to view the app.
  9. The app is best viewed by enabling the Chrome device toolbar and setting the device to iPhone 6/7/8.

Complete Application Setup

Getting Started

  1. Clone this repository by running the following command in a terminal: git clone [email protected]:HACC2018/ohia.ai.git.
  2. Change into the ohia.ai project directory: cd ohia.ai.
  3. When running anything related to the client or server, we require Python 2.7 to be installed, since that is the version that the tfjs-node package supports. Switch your terminal to use Python 2.7 (more information here).
  4. Switch your terminal to use Node 10.12.0 (more information here).

Installation and Execution

  1. Follow the instructions on the API Server README.
  2. Follow the instructions on the App Client README.

Data Collection

Machine learning involves learning from data, thus data collection is an integral part of our application. For more details see ohia.ai/data_collection. View the scraped data in our Google Drive, within the images folder, and our plant metadata within the plant_meta folder.

Machine Learning

Behind the scenes, ohia.ai is powered by machine learning and AI. Our team utilizes modern deep learning techniques and large open source datasets to achieve highly accurate classification on a wide range of flora found throughout the Hawaiian Islands. For more details see ohia.ai/machine_learning.

Results

Plant Name Top 1 Accuracy Top 3 Accuracy Top 5 Accuracy Image Count
Abutilon 80.55% 83.33% 88.88% 374
Achyranthes 53.12% 65.62% 78.12% 258
Aleurites 46.66% 80.00% 86.66% 176
Ananas 86.00% 90.00% 96.00% 612
Anthurium 73.91% 97.10% 97.10% 758
Artocarpus 73.07% 88.46% 88.46% 303
Bonamia 66.66% 66.66% 66.66% 44
Bougainvillea 96.36% 100.00% 100.00% 663
Canavalia 62.50% 79.16% 83.33% 212
Chrysodracon 63.63% 81.81% 81.81% 276
Cocos 83.87% 90.32% 93.54% 311
Colubrina 80.00% 100.00% 100.00% 67
Cordia 36.00% 64.00% 72.00% 209
Cordyline 82.75% 100.00% 100.00% 335
Cortaderia 50.00% 100.00% 100.00% 39
Cyperus 78.57% 91.83% 94.89% 871
Delairea 71.42% 100.00% 100.00% 111
Dioscorea 57.14% 80.00% 85.71% 261
Gouania 66.66% 66.66% 66.66% 31
Heliconia 87.71% 94.73% 98.24% 622
Hibiscus 77.82% 91.63% 95.39% 2,151
Ischaemum 50.00% 75.00% 75.00% 41
Isodendrion 00.00% 50.00% 50.00% 20
Marsilea 75.00% 100.00% 100.00% 51
Mezoneuron 100.00% 100.00% 100.00% 18
Nothocestrum 37.50% 50.00% 75.00% 64
Officinale 66.00% 82.00% 88.00% 556
Panicum 58.06% 90.32% 98.38% 580
Pennisetum 83.33% 100.00% 100.00% 162
Peucedanum 00.00% 50.00% 50.00% 31
Plumeria 79.20% 89.10% 91.08% 741
Portulaca 61.84% 82.89% 82.89% 623
Pritchardia 84.61% 96.15% 98.07% 464
Pseudognaphalium 66.66% 85.18% 94.44% 420
Rubus 84.56% 93.95% 97.31% 1,113
Scaevola 70.66% 84.00% 90.66% 689
Schenkia 12.50% 50.00% 62.50% 48
Senecio 80.80% 92.92% 94.94% 854
Sesbania 60.00% 80.00% 80.00% 104
Tetramolopium 65.00% 90.00% 95.00% 146
Tibouchina 33.33% 100.00% 100.00% 26
Vigna 50.00% 68.75% 75.00% 101

Usage

Find info on the plants of Hawai'i.

Easy-to-follow steps.

AI produces top 3 most likely predictions of plant identity.

An example of most probable plant identity.

Detailed Hawai'i-related information on native plants, including endangerment status.

Historical uses of plants and gallery of uploaded images.

Confirm the plant's identity.

Sponsors

HACC Sponsors Partners