Skip to main content

Week 6

Milestones​

  • Experiment 1: Visual Inspection / Comparison of Tesseract OCR versus new CLIP approach on classifying language
  • Experiment 2:
    • prompt tuning - found best prompt - "image of odiya/english language text"
    • test other CLIP models - best model found - ViT-B/16
    • try setting a threshold parameter that is learnt automatically on the dataset

Screenshots / Videos​

  • image of results of CoOp on the handcrafted dataset comprising 1000+ images image of results of CoOp on the handcrafted dataset comprising 1000+ images

Contributions​

Learnings​

  • Learnt to use CLIP as an effective tool for Zero-Shot image-text tasks.