📄
Sunbird Saral
  • Sunbird Saral Overview
  • Saral Quick Guide
  • Saral Implementation Manual
    • OMR led scanning - Assessments
    • OCR led scanning - Admissions
  • Saral Transformation Story
  • LEARN
    • Software Requirement
    • Application Architecture
    • Features
      • Configurable Branding
      • Capture AI/ML Training Data
      • Support
      • Share App data
      • Auto Sync
      • Multi-Page support
      • Profile Menu
      • Dynamic Validations
      • Dynamic Tagging
      • Minimal Mode
      • Offline mode
      • App Force Update
      • Review results/marks
      • Firebase Analytics and Crashlytics
      • ML model deployment using Firebase
      • Improved Low light Performance - Manual Edit
      • Vertical Forms Scanning Support
      • Improve Processing Speed for big layouts
      • Admissions Data Capture
      • Securing PII Data Capture - Admissions
    • Specifications
      • Layout specification
      • Backend API Swagger Doc
    • Videos
      • Feature Explanation
        • OMR Layout scanning
        • Auto-Sync
        • Share scan app data
        • Skip feature
        • Support feature
        • Validation feature
        • Incorrect scanning
        • Multi-page feature
        • Branding feature
        • Offline mode
        • Review results/marks
      • Usage by States
        • Gujarat Implementation - Between 39:00 - 40:00 mins
        • Uttar Pradesh(U.P) , Gorakhpur Implementation
    • ML Model Accuracy/Results
  • USE
    • Roadmap
    • Workspace Setup - Playbook
    • Saral App Reference Backend
    • Generating APK from source code
    • Generate AAB(App bundle) from source code
    • Sign already generated APK file with private Key
    • Layout configuration
    • Debug/Run Saral App from Android Studio
    • Saral App Debug Tips
    • Saral App Usage Guidelines
    • Update BASE_URL,apkURL in APK
    • Update BASE_URL,apkURL in AAB
    • Sign already generated AAB(Android App Bundle) file with private key
    • Google Play Store App Publish Considerations
    • Layout Design Guidelines
    • Saral OCR Assets
    • Firebase setup for Saral App Telemetry
    • Firebase setup for TFLite model deployment
    • Alternatives for Saral components
  • ENGAGE
    • Source Code Repository
    • Saral SDK Source Code Repository
    • Tracker
    • Releases
      • v1.0.0-beta.1
      • v1.0.0-beta.2
      • v1.0.0-beta.3
      • v1.0.0-beta.4
      • v1.0.0-beta5
      • v1.5.0
      • v1.5.1
      • v1.5.2
      • v1.5.3
      • v1.5.4
      • v1.5.5
      • v1.5.6
      • v1.5.7
      • v1.5.9
      • v1.6.0
      • v1.6.1
      • v1.6.2
      • v1.7.0
    • Saral - Solution Providers
    • Discuss
  • Experience Saral
  • Dev Environment - Installation & Maintenance
    • Saral Installation Guide (Non-Prod)
    • Saral - Sandbox Maintenance Guide (Non-Prod)
  • Saral Easy Installer
    • Saral Production-Environment User Installation Guide
      • Manual Installation for Prod
      • Automating the Infra provisioning and install of the Saral application
        • Prerequisites
        • What automation does
        • Run installer
        • Post install steps
        • Monitoring-Stack
    • Reference Documents
      • SARAL Infra Requirements & Associated Cost
      • Saral Infra Cost Benefit Analysis
  • Tool for Saral Easy Layout generation and Auto generate ROI json
Powered by GitBook
On this page
  • Handwritten Alphanumeric model
  • Handwritten Digits model
  • Sample dataset images

Was this helpful?

Edit on GitHub
  1. LEARN

ML Model Accuracy/Results

Handwritten Alphanumeric model

Iteration 1:

Model details
Dataset
Accuracy achieved

Model trained and inference on

Existing dataset

99.80%

Model inference on

NIST dataset

62.70%

Average across all datasets

-

81.25%

Reasoning - As model is trained on the existing dataset, doesn't perform good on NIST dataset

Iteration 2:

Model details
Dataset
Accuracy achieved

Model trained on

Existing dataset + NIST misclassifications

83.30%

Reasoning - After training the model on NIST misclassifications, improvement in accuracy

Iteration 3:

Model details
Dataset
Accuracy achieved

Model trained on

Existing dataset + NIST misclassifications

93.90%

Reasoning - After training the model on NIST misclassifications as per previous checkpoint, improvement in accuracy

Iteration 4:

Model details
Dataset
Accuracy achieved

Model trained on

Existing dataset + manually collected dataset from sheets (~1K samples)

93.90%

Reasoning - Training the model upon previous checkpoint and adding manually collected data, improvement in accuracy

Handwritten Digits model

Iteration 1:

Model details
Dataset
Accuracy achieved

Model trained and inference on

Existing dataset

99.90%

Model inference on

NIST dataset

60.00%

Model inference on

Obtained production data (~50 samples)

97.70%

Average across all datasets

-

85.80%

Reasoning - As model is trained on the existing dataset, doesn't perform good on NIST dataset

Iteration 2:

Model details
Dataset
Accuracy achieved

Model trained on

Existing dataset + NIST misclasifications

96.40%

Reasoning - After training the model on NIST misclassifications, improvement in accuracy

Iteration 3:

Model details
Dataset
Accuracy achieved

Model trained on

Existing dataset + NIST misclasifications + production dataset (~50 samples)

99.70%

Reasoning: After training the model on NIST misclassifications and production dataset, improvement in accuracy

Iteration 4:

Model details
Dataset
Accuracy achieved

Model trained on

Existing dataset + NIST misclasifications + manually collected dataset from sheets (~8.6k samples)

98.30%

Reasoning: As averaging upon a large production dataset, the accuracy slightly dips as compared to iteration 3

Sample dataset images

Existing dataset

Handwritten alphanumeric

Handwritten digits

NIST dataset

Handwritten alphanumeric

Handwritten digits

Manually collected

Handwritten alphanumeric

Handwritten digits

Some unhandled misclassifications

Reasoning: Generally occurs if the digits are written in corners of the cell

PreviousUsage by StatesNextRoadmap

Last updated 2 years ago

Was this helpful?

0a56f9eb1545428f8d9aea0665b7c460
0b7f1f188faf453b83704a069791f28f
0b0de241b7e64e918fada2f896efdf24
0ac132e3693c47ccb2fe40fb465ba060
00b6ef24818b4323b4ada81ad433af67
0bd15fe615db458781bce7dc32a926b5
0adb585eae814b62a56695650cac0666
0aeaa24048ff4c5d8312d37671c6e08e
0b1a2b29b0904989a3f6df3a13b93d89
0ac2ad40-7af8-44b4-9829-284a2b33416c_printed
0afd4ac792f14b8185d54cb9d0937b3e
0b2c4c540a7247629e1dd7a9abd59962
0abb9b222a1142bdaabd30b6742ca8b2
0b0ffa9c3dcf497c8da259d391f5f159
0aa35cc5-4472-455d-a575-7167c15df849_generated
00af4fd32ddc42efbfb1222725599ecf
0a8986b29e804002a3136ddf110f3638
00c9de07f7b640d083680a5c21661a8b
0__0040c46a-eae7-4219-bffb-7dd418ad9ffb_up_govt
0a1d909f2f2847b795ac40cefa452c3e
hsf_0_00017
hsf_0_00043
hsf_0_00017
hsf_0_00010
hsf_0_00008
hsf_0_00020
hsf_0_00020
hsf_0_00017
hsf_0_00025
hsf_0_00015
hsf_0_00009
hsf_0_00010
hsf_0_00010
hsf_0_00117
hsf_0_00017
hsf_0_00012
img014-032
hsf_0_00009
hsf_0_00019
hsf_0_00009
img026-001
hsf_0_00013
img016-045
img015-044
img023-039
img029-045
img032-053
img013-040
33679
img011-033
33665
33758
img034-055
33753
33745
33362
33650
31075
31372
28926
18103
30658
31065
32358
24141
607
402
32225
24158
30838