A package to get an image and an annotation from a Flutter app, add to the dataset, retrain the model and send the model with the annotation file back.
Add these files to your repository's root directory as given in the folder linked to by the following URL -
https://drive.google.com/drive/folders/1JNZan6HIYXizvNvj-MMpcC5fmkegDfAg?usp=sharing
If the folder is empty, request SR42-dev for the contents.
- loop - continually scan for files in firebase bucket
- download from firebase into local
- delete in firebase
- send file to correct sub directory in dataset
- retrain the fishnofish and classification models with new data
- push model directory content to firebase
The first character of every file uploaded to the /misclassifications bucket on firebase must be any number from 0 to 8 as per the following mapping ...
'0' : 'Gilt-Head Bream'
'1' : 'Black Sea Sprat'
'2' : 'Trout'
'3' : 'Shrimp'
'4' : 'Red Sea Bream'
'5' : 'Striped Red Mullet'
'6' : 'Red Mullet'
'7' : 'Sea Bass'
'8' : 'Hourse Mackerel'
Point to note - This package is only compatible with Ubuntu 20.04 and is not supported as of 26th-August-2022.
Assuming the cloud server repository is public, execute the following commands in the linux terminal once :
git clone https://github.com/SIH-ClapForKrishna/ec2-fishook-pipeline.git
cd ec2-fishook-pipeline
chmod 777 setup
./setup
Alternatively, if you are conversant in ssh, scp & linux kernal programming, feel free to modify and use the included instance-init executable to automate the process of moving this pipeline package to any cloud instance of your choosing.
Once the previous commands are executed, run this script whenever the server has to carry out the aforementioned workflow. Feel free to call this using any other existing script :
./run
To run the pipeline in an infinite loop ...
./server
Keyboard Interrupt for watch :
Ctrl + C