Classification-of-plants-of-Southeast-Asia

This is the 3rd International Competition in Data Science & Artificial Intelligence, organized by The International Society of Data Scientists (ISODS).

Overview

About ISODS

The International Society of Data Scientists (or The ISODS), registered as a Massachusetts non-profit, is a professional organization for Data Science and AI practitioners and researchers, which may include Data Scientists, Machine Learning Scientists, AI Scientists, Data Analysts, Data Engineers, Software Engineers, Risk Analysts, Actuaries, Business Analysts, and more, who apply Data Science and AI at work. ISODS promotes Data Science and AI domestically in the United States as well as internationally via activities such as competitions, conferences, training, professional exams, and publications.

About track: Link

Classification of plants of Southeast Asia is a track in the competition with Bali26 dataset. Bali26 is an image dataset dedicated to ethnobotany, the study of the interaction between people and plants. Bali26 is the first machine-vision ready image collection of ethnobotanically significant flora of south-east Asia, collected on the island of Bali in 2020 (and amended in 2021) together with Balinese residents with intimate local knowledge in coordination with expertise from the Indonesia National Research and Innovation Agency.

Results on Kaggle

Before balancing data

Models	Score	Params
Fine-tuned VGG16 (from layer 15)	0.93339	13,589,402
Fine-tuned VGG16 (all layers)	0.91695	21,223,898
Fine-tuned Resnet50 (all layers)	0.92387	24,065,818
Fine-tuned InceptionResNetV2 (from layer 618)	0.94809	24,342,298
Fine-tuned InceptionResNetV2 (all layers)	0.80968	55,076,474
Fine-tuned InceptionV3 (all layers)	0.72923	22,830,778
Fine-tuned DenseNet201 (all layers)	0.76211	19,089,818

After balancing data

Models	Score	Params	Private Score
Fine-tuned VGG16 (from layer 15)	0.99826	13,589,402
Fine-tuned VGG16 (all layers)	0.99913	21,223,898
Fine-tuned Resnet50 (all layers)	0.99826	24,065,818
Retrain Resnet50 (Don't use pretrain)	0.99134	23,587,866
Fine-tuned InceptionResNetV2 (from layer 618)	0.99913	24,342,298
Retrain InceptionResNetV2 (Don't use pretrain)	1.00000	54,316,154	0.99971 ( rank 12 / 36 )
Fine-tuned InceptionV3 (all layers)	0.98269	22,830,778
Fine-tuned DenseNet201 (all layers)	0.93944	19,089,818
Retrain ResNet101V2	0.99567	42,582,170

How to run

• Git clone:

  git clone https://github.com/htdung167/Classification-of-plants-of-Southeast-Asia.git

• cd:

  cd Classification-of-plants-of-Southeast-Asia/flask_api

1. First way:

• Create 'saved_models' folder in flask_api.
• Download trained model from drive, and put it in saved_models.
• Create env with conda:

  conda create -n <env_name> python=3.7
  conda activate <env_name>
  conda install pip
  pip install -r requirements.txt

• Run:

  python serve.py

2. Or using docker to easily run project:

• Build docker image:

  docker build --tag <image_name>:<image_tag> .

• Run docker container:

  docker run -d -p 5000:5000 --name <container_name> <image_name>:<image_tag>

API Calling

• URL: /predict
• Method: POST
• URL Params: image:[file]
• Success Respone:
  
  • Code: 200
    
  • Content: { "success" : true , "result" : { <10 predicts with the highest probability> } }
• Error Respone:
  
  • Code: 200
    
  • Content: { "success" : false }
• Sample with Postman: