Skip to content

Fix descriptor shape mismatch for LightGlue input #115

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lcmrl merged 1 commit into from
Jan 31, 2026
Merged

Fix descriptor shape mismatch for LightGlue input #115

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
73 changes: 52 additions & 21 deletions src/deep_image_matching/matchers/lightglue.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,31 +5,62 @@
from .matcher_base import FeaturesDict, MatcherBase


def featuresDict2Lightglue(feats: FeaturesDict, device: torch.device) -> dict:
# Remove elements from list/tuple
def featuresDict2Lightglue(feats: dict, device: torch.device) -> dict:
"""
Convert a feature dictionary to LightGlue-compatible format.

Ensures:
keypoints : (B, N, 2)
descriptors : (B, N, D)
scores : (B, N) (if present)
"""

# 1. Unwrap list / tuple
feats = {k: v[0] if isinstance(v, (list, tuple)) else v for k, v in feats.items()}
# Move descriptors dimension to last
if "descriptors" in feats.keys():
if feats["descriptors"].shape[-1] != 256:
feats["descriptors"] = feats["descriptors"].T

if "feature_path" in feats.keys():
del feats["feature_path"]
if "im_path" in feats.keys():
del feats["im_path"]
# if "time" in feats.keys():
# del feats["time"]

# Add batch dimension

# 2. Sanity checks
if "keypoints" not in feats or "descriptors" not in feats:
raise KeyError("features must contain 'keypoints' and 'descriptors'")

kpts = feats["keypoints"] # (N, 2)
desc = feats["descriptors"] # (N, D) or (D, N)

if kpts.ndim != 2 or kpts.shape[1] != 2:
raise ValueError(f"Invalid keypoints shape: {kpts.shape}")

N = kpts.shape[0]

# 3. Fix descriptor layout using keypoint count
if desc.ndim != 2:
raise ValueError(f"Invalid descriptors shape: {desc.shape}")

# Case A: (D, N) → transpose
if desc.shape[1] == N and desc.shape[0] != N:
desc = desc.T

# Case B: (N, D) → OK
elif desc.shape[0] == N:
pass

else:
raise ValueError(
f"Descriptor / keypoint mismatch: "
f"descriptors={desc.shape}, keypoints={kpts.shape}"
)

feats["descriptors"] = desc # now guaranteed (N, D)

# 4. Remove unused keys (LightGlue don't use)
feats.pop("feature_path", None)
feats.pop("im_path", None)

# 5. Add batch dimension
feats = {k: v[None] for k, v in feats.items()}

# Convert to tensor
feats = {
k: torch.tensor(v, dtype=torch.float, device=device) for k, v in feats.items()
}
# Check device
# 6. Convert to torch.Tensor
feats = {
k: v.to(device) if isinstance(v, torch.Tensor) else v for k, v in feats.items()
k: torch.as_tensor(v, dtype=torch.float32, device=device)
for k, v in feats.items()
}

return feats
Expand Down