add: xfeat: https://github.com/verlab/accelerated_features

common/config.yaml

@@ -28,6 +28,10 @@ matcher_zoo:
   aspanformer:
     matcher: aspanformer
     dense: true
+  xfeat:
+    matcher: NN-mutual
+    feature: xfeat
+    dense: false
   dedode:
     matcher: Dual-Softmax
     feature: dedode

common/utils.py

@@ -18,6 +18,8 @@ from .viz import (
     display_matches,
     plot_color_line_matches,
 )
+import time
+import matplotlib.pyplot as plt
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
@@ -462,7 +464,7 @@ def run_matching(
     # update match config
     match_conf["model"]["match_threshold"] = match_threshold
     match_conf["model"]["max_keypoints"] = extract_max_keypoints
-
+    t1 = time.time()
     matcher = get_model(match_conf)
     if model["dense"]:
         pred = match_dense.match_images(
@@ -534,9 +536,9 @@ def run_matching(
         {"geom_info": geom_info},
         choice_estimate_geom,
     )
-
+    plt.close("all")
     del pred
-
+    logger.info(f"TOTAL time: {time.time()-t1:.3f}s")
     return (
         output_keypoints,
         output_matches_raw,

common/viz.py

@@ -252,6 +252,7 @@ def draw_matches(
     img1: np.ndarray,
     conf: np.ndarray,
     titles: Optional[List[str]] = None,
+    texts: Optional[List[str]] = None,
     dpi: int = 150,
     path: Optional[str] = None,
     pad: float = 0.5,
@@ -370,7 +371,10 @@ def draw_image_pairs(
 
 
 def display_matches(
-    pred: Dict[str, np.ndarray], titles: List[str] = [], dpi: int = 300
+    pred: Dict[str, np.ndarray],
+    titles: List[str] = [],
+    texts: List[str] = [],
+    dpi: int = 300,
 ) -> Tuple[np.ndarray, int]:
     """
     Displays the matches between two images.
@@ -408,6 +412,7 @@ def display_matches(
             mconf,
             dpi=dpi,
             titles=titles,
+            texts=texts,
         )
         fig = fig_mkpts
     if (

hloc/extract_features.py

@@ -201,6 +201,17 @@ confs = {
             "resize_max": 1600,
         },
     },
+    "xfeat": {
+        "output": "feats-xfeat-n5000-r1600",
+        "model": {
+            "name": "xfeat",
+            "max_keypoints": 5000,
+        },
+        "preprocessing": {
+            "grayscale": False,
+            "resize_max": 1600,
+        },
+    },    
     "alike": {
         "output": "feats-alike-n5000-r1600",
         "model": {

hloc/extractors/disk.py

@@ -1,5 +1,5 @@
 import kornia
-
+from hloc import logger
 from ..utils.base_model import BaseModel
 
 
@@ -15,6 +15,7 @@ class DISK(BaseModel):
 
     def _init(self, conf):
         self.model = kornia.feature.DISK.from_pretrained(conf["weights"])
+        logger.info(f"Load DISK model done.")
 
     def _forward(self, data):
         image = data["image"]

hloc/extractors/xfeat.py

@@ -0,0 +1,32 @@
+import torch
+from pathlib import Path
+from hloc import logger
+from ..utils.base_model import BaseModel
+
+
+class XFeat(BaseModel):
+    default_conf = {
+        "keypoint_threshold": 0.005,
+        "max_keypoints": -1,
+    }
+    required_inputs = ["image"]
+
+    def _init(self, conf):
+        self.net = torch.hub.load(
+            "verlab/accelerated_features",
+            "XFeat",
+            pretrained=True,
+            top_k=self.conf["max_keypoints"],
+        )
+        logger.info(f"Load XFeat model done.")
+
+    def _forward(self, data):
+        pred = self.net.detectAndCompute(
+            data["image"], top_k=self.conf["max_keypoints"]
+        )[0]
+        pred = {
+            "keypoints": pred["keypoints"][None],
+            "scores": pred["scores"][None],
+            "descriptors": pred["descriptors"].T[None],
+        }
+        return pred