| # Progress Update |
|
|
| --- |
|
|
| ## Project A — Synthetic Test Data Generation |
|
|
| [Project A Overview](project_a_output_progress/project_a_overview.md) |
|
|
| ### What's Done |
|
|
| **Signature Pasting Pipeline** |
| Signatures from CEDAR and SigNet are cleaned, augmented, and pasted onto document templates. Signature field coordinates are currently hardcoded — obtained by running `pdfplumber` on the original born-digital PDF templates and converting PDF point coordinates to PNG pixel coordinates. |
|
|
| > [Document Template](project_a_output_progress/templates) |
|
|
| > [View Synthetic Dataset Output](project_a_output_progress/synthetic) |
|
|
| --- |
|
|
| **Field Auto-Detection** *(in progress)* |
| Currently working on automatically detecting signature and other fields from the template images using OpenCV, so the pipeline works even when new templates are added without needing to manually measure coordinates. |
|
|
| > [View Field Validation Overlays](project_a_output_progress/validation_overlays) |
|
|
| --- |
|
|
| **Test Cases** *(in progress)* |
| 50 structured test cases generated across 5 scenario types (PASS, NO_SIGNATURE, WRONG_PERSON, WRONG_ROLE, REQUIRED_PERSON_MISSING). Each case folder contains the generated document, reference signature images, and a `ground_truth.json`. |
|
|
| > [View Test Cases Output](project_a_output_progress/test_cases) |
