Skip to content

fix: IoU-based instance matching #93

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
josafatburmeister merged 2 commits into from
Apr 19, 2026
Merged

fix: IoU-based instance matching #93

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0613c7d
fix: IoU-based instance matching
josafatburmeister Apr 19, 2026
097f442
fix code formatting
josafatburmeister Apr 19, 2026
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
13 changes: 10 additions & 3 deletions src/pointtorch/metrics/instance_segmentation/_match_instances.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -320,6 +320,7 @@ def match_instances_iou( # pylint: disable=too-many-statements, too-many-locals
| :math:`P` = number of predicted instances
| :math:`T` = number of target instances
"""
print("invalid_instance_id", invalid_instance_id)

if min_iou_treshold is None:
min_iou_treshold = -1
Expand Down Expand Up @@ -360,10 +361,18 @@ def match_instances_iou( # pylint: disable=too-many-statements, too-many-locals

target_ids_to_match, target_batch_indices = torch.unique_consecutive(paired_target_ids, return_inverse=True)

tp, best_predicted_ids = scatter_max(pair_counts, target_batch_indices, dim=0)
pair_target_sizes = target_sizes[paired_target_ids]
pair_predicted_sizes = predicted_sizes[paired_predicted_ids]
pair_ious = pair_counts.to(torch.float) / (
pair_target_sizes.to(torch.float) + pair_predicted_sizes.to(torch.float) - pair_counts.to(torch.float)
)

iou, best_predicted_ids = scatter_max(pair_ious, target_batch_indices, dim=0)
predicted_ids_to_match = paired_predicted_ids[best_predicted_ids]
tp = pair_counts[best_predicted_ids]

has_overlap = tp > 0
iou = iou[has_overlap]
tp = tp[has_overlap]
target_ids_to_match = target_ids_to_match[has_overlap]
predicted_ids_to_match = predicted_ids_to_match[has_overlap]
Expand All @@ -374,8 +383,6 @@ def match_instances_iou( # pylint: disable=too-many-statements, too-many-locals
fp = predicted_sizes - tp
fn = target_sizes - tp

iou = tp.to(torch.float) / (tp.to(torch.float) + fp + fn)

if accept_equal_iou:
matching_mask = iou >= min_iou_treshold
else:
Expand Down
12 changes: 8 additions & 4 deletions test/metrics/instance_segmentation/test_match_instances.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -239,19 +239,23 @@ def test_prediction_with_multiple_matches( # pylint: disable=too-many-locals
):
start_instance_id = invalid_instance_id + 1

target = torch.tensor([0, -1, 2, 2, 3, -1, -1, 1, 1, 1], device=device) + start_instance_id
target = (
torch.tensor([0, -1, 2, 2, 3, -1, -1, 1, 1, 1, 1, -1, -1, -1, -1, -1], device=device) + start_instance_id
)
unique_target_ids = torch.unique(target)
unique_target_ids = unique_target_ids[unique_target_ids != invalid_instance_id]
prediction = torch.tensor([0, 0, 0, 0, 0, -1, -1, 2, 2, 1], device=device) + start_instance_id
prediction = (
torch.tensor([0, 0, 0, 0, 0, -1, -1, 1, 1, 2, -1, 1, 1, 1, 1, 1], device=device) + start_instance_id
)
unique_prediction_ids = torch.unique(prediction)
unique_prediction_ids = unique_prediction_ids[unique_prediction_ids != invalid_instance_id]

expected_matched_target_ids = np.array([2, -1, 1], dtype=np.int64) + start_instance_id
expected_matched_predicted_ids = np.array([0, 2, 0, 0], dtype=np.int64) + start_instance_id
expected_metrics = {
"tp": np.array([1, 2, 2, 1], dtype=np.int64),
"tp": np.array([1, 1, 2, 1], dtype=np.int64),
"fp": np.array([4, 0, 3, 4], dtype=np.int64),
"fn": np.array([0, 1, 0, 0], dtype=np.int64),
"fn": np.array([0, 3, 0, 0], dtype=np.int64),
}

matched_target_ids, matched_predicted_ids, metrics = match_instances_iou(
Expand Down
Loading