Visitar URL original
Fix Faster R-CNN detector inference crashing on box_score_thresh: null by rysk-t · Pull Request #3470 · DeepLabCut/DeepLabCut · GitHub
Skip to content

Fix Faster R-CNN detector inference crashing on box_score_thresh: null #3470

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
deruyter92 merged 14 commits into from
Sep 3, 2026
Merged

Fix Faster R-CNN detector inference crashing on box_score_thresh: null #3470

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
fcfb6eb
Treat a null box_score_thresh as the default detector threshold
rysk-t Aug 30, 2026
6db842d
Normalize detector box score threshold
C-Achard Aug 31, 2026
21a0eb9
Refactor detector threshold tests to config level
C-Achard Aug 31, 2026
c61ddcc
Fix incorrect import
C-Achard Aug 31, 2026
aac861c
Fix model config test defaults for TD
C-Achard Sep 1, 2026
85072d9
Use config cutoff for detector boxes
C-Achard Sep 1, 2026
7893a35
Fix detector box_score_thresh handling
C-Achard Sep 1, 2026
5e58815
Clarify detector threshold docs
C-Achard Sep 1, 2026
3e2611e
Fix `create_video_with_all_detections` -> read default `bbox_pcutoff`…
deruyter92 Sep 1, 2026
baff692
revert shared box score threshold default -> inline config default
deruyter92 Sep 1, 2026
0f5330a
update docstring for `DetectorModelConfig`
deruyter92 Sep 1, 2026
256dfb9
add test for building config with null threshold
deruyter92 Sep 1, 2026
2a00c40
Make `ProjectConfig.bboxes_pcutoff`` a property rather than a field
deruyter92 Sep 1, 2026
2689d30
add xfail for `ProjectConfig.bboxes_pcutoff`
deruyter92 Sep 1, 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
Prev Previous commit
Next Next commit
Normalize detector box score threshold 
Set a consistent default for `box_score_thresh` and normalize `None` values before model construction. `DetectorModelConfig` now defaults to `0.01` and uses a validator to translate `null` config values to that default, while the torchvision detector adaptor reuses the shared constant instead of hardcoded literals. This prevents Faster R-CNN inference crashes caused by `None` threshold comparisons.
  • Loading branch information
@C-Achard
C-Achard committed Aug 31, 2026
commit 6db842dbaa4f5497aa18e864cebe041aa334a61e
11 changes: 9 additions & 2 deletions deeplabcut/pose_estimation_pytorch/config/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,11 +10,13 @@
#
"""Model configuration class for DeepLabCut pose estimation models."""

from pydantic import Field
from pydantic import Field, field_validator

from deeplabcut.core.config import DLCBaseConfig
from deeplabcut.core.config.validation import Fraction

DEFAULT_BOX_SCORE_THRESH = 0.01


class ModelConfig(DLCBaseConfig):
"""Complete model configuration.
Expand Down Expand Up @@ -48,4 +50,9 @@ class DetectorModelConfig(DLCBaseConfig):
freeze_bn_stats: bool = False
freeze_bn_weights: bool = False
variant: str | None = None
box_score_thresh: Fraction | None = None
box_score_thresh: Fraction | None = DEFAULT_BOX_SCORE_THRESH

@field_validator("box_score_thresh", mode="before")
@classmethod
def normalize_box_score_thresh(cls, v):
return DEFAULT_BOX_SCORE_THRESH if v is None else v
5 changes: 3 additions & 2 deletions deeplabcut/pose_estimation_pytorch/models/detectors/torchvision.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
import torch
import torchvision.models.detection as detection

from deeplabcut.pose_estimation_pytorch.config.pose import DEFAULT_BOX_SCORE_THRESH
from deeplabcut.pose_estimation_pytorch.models.detectors.base import (
BaseDetector,
)
Expand Down Expand Up @@ -58,7 +59,7 @@ def __init__(
num_classes: int | None = 2,
freeze_bn_stats: bool = False,
freeze_bn_weights: bool = False,
box_score_thresh: float | None = 0.01,
box_score_thresh: float | None = DEFAULT_BOX_SCORE_THRESH,
model_kwargs: dict | None = None,
) -> None:
super().__init__(
Expand All @@ -71,7 +72,7 @@ def __init__(
# Generated configs store ``box_score_thresh: null`` meaning "use
# the default"; torchvision's Faster R-CNN would keep the None and
# crash when comparing scores against it during inference.
box_score_thresh = 0.01
box_score_thresh = DEFAULT_BOX_SCORE_THRESH

# Load the model
model_fn = getattr(detection, model)
Expand Down