drywall-grounded-sam
Two-stage prompted segmentation: Grounding DINO (grounding-dino-tiny) emits
bboxes from a text prompt โ SAM (sam-vit-base) converts boxes โ masks โ union.
Both stages fine-tuned on the unified drywall pool (3 epochs each).
Test metrics
| Task | Dice | mIoU | Precision | Recall |
|---|---|---|---|---|
| Crack | 0.601 | 0.463 | 0.637 | 0.725 |
| Taping | 0.515 | 0.393 | 0.491 | 0.703 |
Load + predict
from huggingface_hub import snapshot_download
local = snapshot_download(repo_id="ravindrakapse/drywall-grounded-sam")
# local/gd_finetuned/ -> AutoProcessor + AutoModelForZeroShotObjectDetection
# local/sam_finetuned/ -> SamProcessor + SamModel
See load_models.py for the full pipeline.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support