SAM2 Household Fine-tuned Checkpoint
Fine-tuned facebook/sam2.1-hiera-small for household object masks with bbox prompts.
- Base model:
facebook/sam2.1-hiera-small - Base revision:
ee5bba1d82bb8749febdf90f45e84b687142ba03 - Dataset source:
COCO 2017 instance segmentation - Dataset revision:
annotations_trainval2017 - Split manifest SHA-256:
509e83782606d6dc387c6c22ef5e04c2ae21a2907526a9df2f3cdde9a8338ae9 - Training samples: 512
- Validation samples: 128
- Test samples: 128
- Steps: 10
- Batch size: 16
- Best validation IoU: 0.742352
- Best validation Dice: 0.727286
- Production output mode: multimask with predicted-IoU candidate selection
- Test selected-candidate IoU: 0.729971
- Test oracle-candidate IoU: 0.758795
- Test selection regret: 0.028824
- Test IoU-head MAE: 0.185972
- W&B: https://wandb.ai/sshibinthomass-tu-dortmund/household-object-finetuning/runs/ax0fxcku
- Modal app:
sam2-household-modal-train-v3 - Hub repo:
sshibinthomass/sam2-household-accuracy-v3-20260710-170002
Usage
from transformers import Sam2Model, Sam2Processor
processor = Sam2Processor.from_pretrained("sshibinthomass/sam2-household-accuracy-v3-20260710-170002")
model = Sam2Model.from_pretrained("sshibinthomass/sam2-household-accuracy-v3-20260710-170002")
outputs = model(**processor(images=image, input_boxes=[[box]], return_tensors="pt"), multimask_output=True)
best_mask_index = outputs.iou_scores[:, 0].argmax(dim=-1)