Progress Reports

Stage I Report: 

Planning & Requirements:

  • Defined problem: lack of holistic and interpretable testing for image classifiers 
  • Identified key components: 
    • Model & dataset understanding 
    • Multiple testing strategies (random, stress, feature-based, AI-guided) 
  • Finalized initial tech stack: 
    • React + Vite (UI), Node.js (API), Python (ML execution), LLM integration 
  • Selected baseline models (e.g., ResNet, VGG) 

Stage II Report: 

Prototype Design & Initial Implementation:

  • Designed system architecture and workflow 
  • Implemented: 
    • Model summary module (LLM-based)  
    • Basic UI for model and dataset input 
  • Developed Random Testing module: 
    • Input label + dataset 
    • Generated pass/fail results and metrics 
  • Initial integration between frontend, backend, and Python inference 

Stage III Report: 

Stress Testing & Analysis:

  • Implemented Stress Testing module: 
    • Background changes 
    • Environment variations 
    • Occlusion and multi-object scenarios 
  • Enabled before vs after comparison of predictions 
  • Observations: 
    • Significant drop in performance under transformations 
    • Confidence scores not reliable for robustness 
  • Added basic reporting for test results and failure counts 

Stage IV Report: 

Current Progress & Ongoing Work:

  • Currently working on Feature based Testing: 
    • Generating visual features using LLM 
    • Designing targeted test cases based on features 
  • Improving: 
    • Failure analysis using LLM 
    • Final report generation (summary + insights) 
  • Next steps: 
    • Complete feature testing pipeline 
    • Integrate all modules into unified reporting