Add phikon-v2 to registry

docs/user-guide/advanced/replicate_evaluations.md

@@ -15,7 +15,7 @@ Keep in mind:
 
 ### DINO ViT-S16 (random weights)
 
-Evaluating the backbone with randomly initialized weights serves as a baseline to compare the pretrained FMs to an FM that produces embeddings without any prior learning on image tasks. To evaluate, run:
+Evaluating the backbone with randomly initialized weights serves as a baseline to compare the pretrained FMs to a FM that produces embeddings without any prior learning on image tasks. To evaluate, run:
 
 ```
 MODEL_NAME="universal/vit_small_patch16_224_random" \
@@ -62,7 +62,7 @@ eva predict_fit --config configs/vision/pathology/offline/<task>.yaml
 
 ### Phikon (Owkin) - iBOT ViT-B16 (TCGA) [[2]](#references)
 
-[Owkin](https://www.owkin.com/) released the weights for "Phikon", an FM trained with iBOT on TCGA data, via
+[Owkin](https://www.owkin.com/) released the weights for "Phikon", a FM trained with iBOT on TCGA data, via
 [HuggingFace](https://huggingface.co/owkin/phikon). To evaluate, run:
 
 ```
@@ -73,6 +73,20 @@ IN_FEATURES=768 \
 eva predict_fit --config configs/vision/pathology/offline/<task>.yaml
 ```
 
+### Phikon-v2 (Owkin) - DINOv2 ViT-L16 (PANCAN-XL) [[9]](#references)
+
+[Owkin](https://www.owkin.com/) released the weights for "Phikon-v2", a FM trained with DINOv2
+on the PANCAN-XL dataset (450M 20x magnification histology images sampled from 60K WSIs), via
+[HuggingFace](https://huggingface.co/owkin/phikon-v2). To evaluate, run:
+
+```
+MODEL_NAME=pathology/owkin_phikon_v2 \
+NORMALIZE_MEAN="[0.485,0.456,0.406]" \
+NORMALIZE_STD="[0.229,0.224,0.225]" \
+IN_FEATURES=1024 \
+eva predict_fit --config configs/vision/pathology/offline/<task>.yaml
+```
+
 ### UNI (MahmoodLab) - DINOv2 ViT-L16 (Mass-100k) [[3]](#references)
 
 The UNI FM by MahmoodLab is available on [HuggingFace](https://huggingface.co/MahmoodLab/UNI). Note that access needs to 
@@ -219,4 +233,6 @@ eva predict_fit --config configs/vision/pathology/offline/<task>.yaml
 
  [6]: Xu, Hanwen, et al. "A whole-slide foundation model for digital pathology from real-world data." Nature (2024): 1-8.
 
- [7]: Nechaev, Dmitry, Alexey Pchelnikov, and Ekaterina Ivanova. "Hibou: A Family of Foundational Vision Transformers for Pathology." arXiv preprint arXiv:2406.05074 (2024).
+ [7]: Nechaev, Dmitry, Alexey Pchelnikov, and Ekaterina Ivanova. "Hibou: A Family of Foundational Vision Transformers for Pathology." arXiv preprint arXiv:2406.05074 (2024).
+
+ [9]: Filiot, Alexandre, et al. "Phikon-v2, A large and public feature extractor for biomarker prediction." arXiv preprint arXiv:2409.09173 (2024).

src/eva/vision/models/networks/backbones/pathology/__init__.py

@@ -12,7 +12,7 @@
 )
 from eva.vision.models.networks.backbones.pathology.lunit import lunit_vits8, lunit_vits16
 from eva.vision.models.networks.backbones.pathology.mahmood import mahmood_uni
-from eva.vision.models.networks.backbones.pathology.owkin import owkin_phikon
+from eva.vision.models.networks.backbones.pathology.owkin import owkin_phikon, owkin_phikon_v2
 
 __all__ = [
     "kaiko_vitb16",
@@ -21,6 +21,7 @@
     "kaiko_vits16",
     "kaiko_vits8",
     "owkin_phikon",
+    "owkin_phikon_v2",
     "lunit_vits16",
     "lunit_vits8",
     "mahmood_uni",

src/eva/vision/models/networks/backbones/pathology/owkin.py

@@ -20,3 +20,17 @@ def owkin_phikon(out_indices: int | Tuple[int, ...] | None = None) -> nn.Module:
         The model instance.
     """
     return _utils.load_hugingface_model(model_name="owkin/phikon", out_indices=out_indices)
+
+
+@register_model("pathology/owkin_phikon_v2")
+def owkin_phikon_v2(out_indices: int | Tuple[int, ...] | None = None) -> nn.Module:
+    """Initializes the phikon-v2 pathology FM by owkin (https://huggingface.co/owkin/phikon-v2).
+
+    Args:
+        out_indices: Whether and which multi-level patch embeddings to return.
+            Currently only out_indices=1 is supported.
+
+    Returns:
+        The model instance.
+    """
+    return _utils.load_hugingface_model(model_name="owkin/phikon-v2", out_indices=out_indices)