update: d2net model links and SIFT

common/utils.py

@@ -5,7 +5,7 @@ import torch
 from itertools import combinations
 import cv2
 import gradio as gr
-from hloc import matchers, extractors
+from hloc import matchers, extractors, logger
 from hloc.utils.base_model import dynamic_load
 from hloc import match_dense, match_features, extract_features
 from hloc.utils.viz import add_text, plot_keypoints
@@ -187,14 +187,17 @@ def compute_geom(
         )
         if H is not None:
             geo_info["Homography"] = H.tolist()
-            _, H1, H2 = cv2.stereoRectifyUncalibrated(
-                mkpts0.reshape(-1, 2),
-                mkpts1.reshape(-1, 2),
-                F,
-                imgSize=(w1, h1),
-            )
-            geo_info["H1"] = H1.tolist()
-            geo_info["H2"] = H2.tolist()
+            try:
+                _, H1, H2 = cv2.stereoRectifyUncalibrated(
+                    mkpts0.reshape(-1, 2),
+                    mkpts1.reshape(-1, 2),
+                    F,
+                    imgSize=(w1, h1),
+                )
+                geo_info["H1"] = H1.tolist()
+                geo_info["H2"] = H2.tolist()
+            except cv2.error as e:
+                logger.error(f"e, skip")
         return geo_info
     else:
         return {}

hloc/extractors/d2net.py

@@ -26,7 +26,7 @@ class D2Net(BaseModel):
             model_file.parent.mkdir(exist_ok=True)
             cmd = [
                 "wget",
-                "https://dsmn.ml/files/d2-net/" + conf["model_name"],
+                "https://dusmanu.com/files/d2-net/" + conf["model_name"],
                 "-O",
                 str(model_file),
             ]

hloc/extractors/dog.py

@@ -44,6 +44,7 @@ class DoG(BaseModel):
             raise ValueError(f'Unknown descriptor: {conf["descriptor"]}')
 
         self.sift = None  # lazily instantiated on the first image
+        self.dummy_param = torch.nn.Parameter(torch.empty(0))
         self.device = torch.device("cpu")
 
     def to(self, *args, **kwargs):
@@ -63,7 +64,8 @@ class DoG(BaseModel):
         assert image_np.min() >= -EPS and image_np.max() <= 1 + EPS
 
         if self.sift is None:
-            use_gpu = pycolmap.has_cuda and self.device.type == "cuda"
+            device = self.dummy_param.device
+            use_gpu = pycolmap.has_cuda and device.type == "cuda"
             options = {**self.conf["options"]}
             if self.conf["descriptor"] == "rootsift":
                 options["normalization"] = pycolmap.Normalization.L1_ROOT
@@ -73,8 +75,7 @@ class DoG(BaseModel):
                 options=pycolmap.SiftExtractionOptions(options),
                 device=getattr(pycolmap.Device, "cuda" if use_gpu else "cpu"),
             )
-
-        keypoints, scores, descriptors = self.sift.extract(image_np)
+        keypoints, descriptors = self.sift.extract(image_np)
         scales = keypoints[:, 2]
         oris = np.rad2deg(keypoints[:, 3])
 
@@ -109,7 +110,8 @@ class DoG(BaseModel):
         keypoints = torch.from_numpy(keypoints[:, :2])  # keep only x, y
         scales = torch.from_numpy(scales)
         oris = torch.from_numpy(oris)
-        scores = torch.from_numpy(scores)
+        scores = keypoints.new_zeros(len(keypoints))  # no scores for SIFT yet
+
         if self.conf["max_keypoints"] != -1:
             # TODO: check that the scores from PyCOLMAP are 100% correct,
             # follow https://github.com/mihaidusmanu/pycolmap/issues/8