As a deployer/Maintainer, I should be able to improve the Handwritten Block letter Alphanumeric model Accuracy with production datasets - Iteration 3

[dileep-gadiraju]

As a deployer/Maintainer, I should be able to improve the Handwritten Block letter Alphanumeric model Accuracy with production datasets - Iteration 3 :

Pre-requisite: Get handwritten layout sheets from the field with different users (considering demographics like gender, geography etc. of the users)

• Scan and collect image data of miss-predicted Handwritten Block Letter Alphanumeric models from the layout sheets filled by various users.
• Generate Synthetic data from production data as needed.
• Prepare complete training data
• Experiment with the existing model version (https://github.com/Sunbird-Saral/Project-Saral/tree/main/v1.0/frontend/SaralApp/android/saralsdk/src/main/assets#:~:text=model_letter_digits_finetune_18_07_22_epoch_15.tflite) to identify the miss-classifications.
• Scan and
• Training the next version of the model.
• Inferencing and identifying optimal model version/checkpoint.
• Package with SaralSDK/SaralReference App to test in real-time.