Fix param counts: DA3-NESTED-GIANT-LARGE is 1.69B, total 2.5B

README.md

@@ -6,8 +6,6 @@ tags:
 - segmentation
 - language-grounding
 - pose-free
-datasets:
-- scannetpp
 pipeline_tag: image-segmentation
 ---
 
@@ -25,25 +23,22 @@ TrianguLang is a feed-forward, pose-free method for language-guided 3D localizat
 
 ## Checkpoints
 
-| Checkpoint | Description | Eval mIoU |
-|---|---|---|
-| `mo_v11/best.pt` | Multi-object (text + spatial), 230 scenes, 8 views, 100 epochs | Best MO |
-| `fullscale_no_qp/best.pt` | Single-object (text-only), 230 scenes, 100 epochs | Best SO |
-
-## Usage
+| Checkpoint | Description |
+|---|---|
+| `mo_v11/best.pt` | Multi-object (text + spatial), 230 scenes, 8 views, 100 epochs |
+| `fullscale_no_qp/best.pt` | Single-object (text-only), 230 scenes, 100 epochs |
 
-```python
-from triangulang.training.train import TrianguLangModel
+## Architecture
 
-checkpoint = torch.load("best.pt", map_location="cpu")
-model = TrianguLangModel(...)
-model.load_state_dict(checkpoint["model_state_dict"])
-```
+- **Frozen:** SAM3 (841M) + DA3-NESTED-GIANT-LARGE (1.69B) = ~2.5B params
+- **Trainable:** GASA Decoder (~13.5M params)
 
-## Architecture
+## Results (ScanNet++)
 
-- **Frozen:** SAM3 (848M) + DA3 (335M) = ~1.2B params
-- **Trainable:** GASA Decoder (~10-12M params)
+| Setting | mIoU | mAcc |
+|---|---|---|
+| Text-only (single-object) | **62.4%** | **77.4%** |
+| Text-only + CRF | **65.2%** | - |
 
 ## Citation