| # Project Status | |
| VintageGAN is currently an **academic research prototype**. | |
| ## Current State | |
| - The target user experience is preset/slider-based vintage filtering. | |
| - The technical goal is controllable conditional image-to-image synthesis. | |
| - The default training profile is `local_256` for RTX 3050 4GB class GPUs. | |
| - Cloud profiles are available for 384px and 512px experimentation. | |
| ## Ready | |
| - Procedural target generation for six vintage defect controls. | |
| - Conditional generator/discriminator scaffolding. | |
| - Explicit preset vectors in `configs/presets.yaml`. | |
| - Local/cloud config profiles in `configs/training_config.yaml`. | |
| - Inference API/CLI contract with optional metadata JSON. | |
| ## Needs Real Validation | |
| - Clean environment installation. | |
| - End-to-end smoke training. | |
| - Real trained checkpoints. | |
| - Real metrics generated from saved experiment artifacts. | |
| - Dataset source/license documentation. | |
| - Detector-backed consistency loss experiments. | |
| ## Not Claimed | |
| - Production readiness. | |
| - Publication readiness. | |
| - Academic-grade results. | |
| - Real FID/SSIM/PSNR numbers. | |
| - Bundled trained checkpoints. | |
| This file should be updated only from verified runs, not from intended targets. | |