OMM

On my mark is a computer vision solution for automated floating debris detection.

---

Models

Preview

Model Name	Model size	Model Characteristics	Recommended Inference Hardware
OG_yolo11n.pt	2.59 M	A mere fine-tuned yolo11n with custom dataset	CPU, GPU, NPU
distilled_yolo11n.pt	2.62 M	META DINOv3 distilled yolo11n	CPU,GPU, NPU
dfine.pt	36.79 M	Trained DFINE detection header for DINOv3	GPU
dfine_fp32.onnx	36.79 M	ONNX format of the DFINE model for acceleration	GPU

model details

Disclaim : All three models was based on Meta's facebook/dinov3-vits16-pretrain-lvd1689m model. Key training/distillation framework is from LightlyTrain

1. distilled_yolo11n.pt [TBC]
2. dfine.pt [TBC]
3. dfine_fp32 [TBC]

---

Inference Backend

All the model can be inferenced with standard python inference framework like pytorch and onnxruntime. For graphical UI. You may reference to: OMM