Skip to content

Fix NaN handling in PredictKeypointIdentities postprocessor #3378

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

Draft
deruyter92 wants to merge 1 commit into
base: main
Choose a base branch
from
+50 −4
Draft

Fix NaN handling in PredictKeypointIdentities postprocessor #3378

Show file tree
Hide file tree
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
12 changes: 8 additions & 4 deletions deeplabcut/pose_estimation_pytorch/data/postprocessor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -504,13 +504,17 @@ def __call__(self, predictions: dict[str, np.ndarray], context: Context) -> tupl

id_score_matrix = np.zeros((num_preds, num_keypoints, num_ids))
for pred_idx, individual_keypoints in enumerate(pose):
heatmap_indices = np.rint(individual_keypoints).astype(int)
xy = individual_keypoints[:, :2]
valid = np.all(np.isfinite(xy), axis=1)

heatmap_indices = np.zeros((num_keypoints, 2), dtype=int)
if np.any(valid):
heatmap_indices[valid] = np.rint(xy[valid]).astype(int)
Comment on lines +510 to +512

@C-Achard C-Achard Jun 22, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could we shorten this by directly using the valid mask on heatmap_indices ?

xs = np.clip(heatmap_indices[:, 0], 0, w - 1)
ys = np.clip(heatmap_indices[:, 1], 0, h - 1)

# get the score from each identity heatmap at each predicted keypoint
for kpt_idx, (x, y) in enumerate(zip(xs, ys, strict=False)):
id_score_matrix[pred_idx, kpt_idx] = identity_heatmap[y, x, :]
if valid[kpt_idx]:
id_score_matrix[pred_idx, kpt_idx] = identity_heatmap[y, x, :]

predictions[self.identity_key] = id_score_matrix
if not self.keep_id_maps:
Expand Down
42 changes: 42 additions & 0 deletions tests/pose_estimation_pytorch/data/test_postprocessor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -379,3 +379,45 @@ def test_remove_low_confidence_boxes(data):

np.testing.assert_array_equal(predictions["bboxes"], expected_bboxes)
np.testing.assert_array_equal(predictions["bbox_scores"], expected_scores)


def test_predict_keypoint_identities_handles_nan_keypoints():
import warnings

p = PredictKeypointIdentities(
identity_key="keypoint_identity",
identity_map_key="identity_map",
pose_key="bodyparts",
keep_id_maps=True,
)

# PAF-style output: (num_individuals, num_bodyparts, 5); missing joint is all-NaN
bodyparts = np.array(
[
[
[3.1, 1.0, 0.8, 0.0, 0.5], # valid
[np.nan, np.nan, np.nan, np.nan, np.nan], # missing (assembler default)
[1.0, 0.0, 0.9, 1.0, 0.5], # valid
],
]
)
id_heatmap = np.array(
[
[[0.1, 0.1], [0.2, 0.1], [0.3, 0.1], [0.4, 0.1]],
[[0.1, 0.2], [0.2, 0.2], [0.3, 0.2], [0.4, 0.2]],
[[0.1, 0.3], [0.2, 0.3], [0.3, 0.3], [0.4, 0.3]],
[[0.1, 0.4], [0.2, 0.4], [0.3, 0.4], [0.4, 0.4]],
]
)
predictions_in = {"bodyparts": bodyparts, "identity_map": id_heatmap}

with warnings.catch_warnings():
warnings.simplefilter("error", RuntimeWarning)
predictions, _ = p(predictions_in, {})

expected = np.zeros((1, 3, 2))
expected[0, 0] = id_heatmap[1, 3] # rint(3.1, 1.0) -> (3, 1)
expected[0, 1] = 0.0 # NaN keypoint: leave identity scores at zero
expected[0, 2] = id_heatmap[0, 1] # rint(1.0, 0.0) -> (1, 0)

np.testing.assert_array_equal(predictions["keypoint_identity"], expected)
Loading