| # Monitoring images — drop dashboard screenshots here | |
| Save the screenshots you capture from the training dashboard **into this folder**, using | |
| these exact filenames so the model card references them automatically. PNG or JPG, any | |
| resolution (higher is better — these become figures in the published model card). | |
| | filename | what to capture | | |
| |---|---| | |
| | `map_curve.png` | The **mAP50-95 curve** over epochs (both the live curve and, if shown, the `ultra_mAP50_95` series). | | |
| | `map50_curve.png` | The **mAP50** curve over epochs. | | |
| | `loss_curves.png` | The **box / cls / dfl loss** curves over epochs. | | |
| | `precision_recall.png` | The **precision & recall** curves over epochs. | | |
| | `gpu_util.png` | The **GPU panel** — 3× RTX 3090 utilization / temperature / power (shows the 200 W cap). | | |
| | `overview.png` | A **full-dashboard** overview screenshot (the run page as a whole). | | |
| Notes | |
| - Any extra images are welcome — add them and tell me the caption; I'll wire them into | |
| the model card. | |
| - These come from **your** screenshots of the dashboard. The dashboard page itself | |
| (`report.html` / the live site) is **never** published — only the images you export. | |
| - Once the images are here, the release-prep step embeds them and the model card's | |
| "Training curves" section will render them. | |