reTraining/ORIGINAL_REQUEST.md

Original User Request

Initial Request — 2026-08-27T05:02:32Z

Run the dataset enrichment & retraining app (FastAPI backend + Vite/React frontend), systematically capture high-resolution screenshots of all pages, modal dialogues, interactive states, and progress workflows into a structured folder, and execute an automated QC agent to verify capture completeness across all workflows.

Working directory: /home/asus/feedmill/reTraining Output directory: /home/asus/feedmill/reTraining/screenshots

Capture Target List

1. Core Pages & Views

1. Projects Page (#/projects)
   • Main projects overview
   • Project creation modal / form
2. Library / Video Archive (#/projects/:id)
   • Cycle & recording session tree
   • Batch selection & status indicators
3. Trim Page (#/projects/:id/trim/:rel)
   • Video trim editor & frame rate extraction configuration
4. Batches Page (#/projects/:id/batches)
   • Batch extraction status & table
   • SAM3 Auto-Annotate modal
   • Mass Auto-Annotate modal
5. Review & Annotation Page (#/projects/:id/review?batch=:id)
   • Annotation canvas (boxes/polygons) & class palette
   • Filmstrip navigation & quick reclass bar
   • Triage crop grid & scatter plot views
   • Exemplar pool & outlier filter panel
6. Data Prep Page (#/projects/:id/data-prep?batches=:ids)
   • Quality filter rules & outlier detection
   • Augmentation configuration panel
   • Merge Target modal (freezing dataset split)
7. Datasets Page (#/projects/:id/datasets)
   • Dataset list, class distribution charts, and train/val split stats
8. Models & Training Page (#/projects/:id/models)
   • Model fine-tuning configuration & live training progress / loss curves
   • Benchmark comparison (Base model vs New model mAP)
9. Counting Bench Page (#/projects/:id/counting-bench)
   • Counting evaluation benchmark against video batches
10. Live Count Page (#/projects/:id/live-count)
    • Real-time video/RTSP inference stream & counting metrics overlay
11. SAM3 Playground (#/sam3-playground)
    • Interactive text prompt grounding playground

2. Workflow Progress & Modal States

• Frame extraction progress state
• SAM3 auto-annotation loading & completion state
• Data preparation filtering & merge confirmation
• Model training & evaluation progress state

Requirements

R1. Application Startup & Environment Verification

• Start FastAPI backend on :8000 and Vite frontend on :5173 (or production port) using uv and npm.
• Ensure all sample data or test project fixtures are accessible.

R2. Automated Screenshot Capture Engine

• Implement an automated capture script using Playwright / Puppeteer / headless browser.
• Navigate to every listed route, trigger interactive states (modals, tabs, triage views, progress indicators), and capture full-screen images.
• Store all images with clear, hierarchical naming in screenshots/ (e.g. 01_projects_page.png, 04_batches_auto_annotate_modal.png, etc.).

R3. Dedicated QC Agent & Verification Report

• Run an independent QC verification step that inspects the generated screenshot directory against the full capture checklist.
• Produce a structured markdown report screenshots/QC_REPORT.md confirming coverage, image clarity, and zero missing workflow states.

Acceptance Criteria

Visual & Workflow Coverage

• All 11 page types captured at minimum 1440x900 resolution.
• All modal dialogs (Create Project, SAM3 Auto-Annotate, Mass Auto-Annotate, Merge Target) captured.
• Review page interactive components (Triage, Filmstrip, Exemplar panel) captured.
• Screenshots saved in dedicated screenshots/ directory with standardized naming.
• Independent QC audit passes 100% of items in the capture checklist.