Add demo reference, encoder feature extraction docs

README.md

@@ -34,8 +34,14 @@ All models use a ViT-Large encoder (24 layers, 1024-dim) and a ViT-Base decoder
 
 ## Usage
 
+A complete demo with visualization is provided in [`demo.py`](https://github.com/thibautloiseau/alligat0r/blob/main/demo.py) on the GitHub repository.
+
+### Covisibility prediction
+
 ```python
 import torch
+from PIL import Image
+from torchvision import transforms
 from reloc3r.alligat0r import Alligat0R
 
 device = "cuda"
@@ -45,9 +51,6 @@ model = Alligat0R.from_pretrained(
     device=device,
 )
 
-from PIL import Image
-from torchvision import transforms
-
 img_size = (384, 512)  # use (288, 512) for nuScenes variants
 tf = transforms.Compose([
     transforms.Resize(img_size),
@@ -64,6 +67,20 @@ cov1 = seg1.argmax(1)[0].cpu().numpy()  # 0=covisible, 1=occluded, 2=outside-FOV
 cov2 = seg2.argmax(1)[0].cpu().numpy()
 ```
 
+### Encoder feature extraction
+
+The pre-trained encoder can be used as a feature backbone for downstream tasks:
+
+```python
+feat1, feat2, pos1, pos2 = model.encode(view1, view2)
+# feat1, feat2: (B, N_patches, 1024)  — ViT-L encoder features
+# pos1, pos2:   (B, N_patches, 2)     — 2-D patch positions
+
+# To get intermediate features from all 24 encoder blocks:
+feat1_all, feat2_all, pos1, pos2 = model.encode(view1, view2, return_all_blocks=True)
+# feat1_all: list of 24 tensors, each (B, N_patches, 1024)
+```
+
 ## Fine-Tuning for Pose Regression
 
 These pre-trained weights serve as initialization for downstream tasks. To fine-tune for relative pose regression, use the training script from the [GitHub repository](https://github.com/thibautloiseau/alligat0r):