ribline-scf-pass2-autocontext: Two-Pass Auto-Context Semantic Context Forest Baseline

Official Pass-1 and Pass-2 Auto-Context Semantic Context Forest (SCF) baseline checkpoints (rib_forest_pass1.joblib, rib_forest_pass2.joblib) and 24-rib statistical atlas (rib_atlas.joblib) for ribline (v0.3.0), described in Hybrid Context Forests and Bilateral Viterbi Matching for 3D Rib Centerline Tracing in CT (PDF) and part of the ribline Hugging Face Collection:

Hugging Face Repository	Model Variant	12-Scan LCD (mm)	12-Scan Acc (%)	5-Fold CV LCD (mm)	5-Fold CV Acc (%)
wq2012/ribline-3d-ct-centerline	Hybrid SCF + 3D U-Net + Stage 3 (Full Suite)	2.05 +/- 0.48	97.2 %	3.64 +/- 3.90	91.8 %
wq2012/ribline-anatomical-scf	Anatomically Anchored SCF + Stage 3 (CPU-only)	2.88 +/- 1.32	95.1 %	5.28 +/- 4.62	86.9 %
wq2012/ribline-3d-unet	Lightweight 3D U-Net + Stage 3	3.40 +/- 1.99	92.0 %	7.44 +/- 8.93	78.2 %
wq2012/ribline-scf-pass2-autocontext	2-Pass Auto-Context SCF Baseline + Stage 3	9.36 +/- 10.57	71.0 %	--	--

---

Model Description

This repository archives the classical Two-Pass Auto-Context Semantic Context Forest ablation baseline (extending Wang et al., MICCAI MCV 2013 from knee MRI cartilage segmentation to 3D thoracic CT rib centerlines):

1. Pass-1 Appearance Forest (rib_forest_pass1.joblib): 120 trees, 228 features comprising raw CT intensity, 27 multi-scale Gaussian derivatives (sigma = 1, 2, 4 mm), and 200 Random Shift Intensity Difference (RSID) pairs with offsets in [-20, 20] voxels (+/-40 mm), gated by an empirical high-intensity bone mask (I_raw > 1150, roughly > 126 HU).
2. Pass-2 Auto-Context Forest (rib_forest_pass2.joblib): 100 trees, 306 features combining the 228 Pass-1 appearance features with 28 multi-scale differential features of the Pass-1 probability map and 50 long-range random-shift probability context differences in [-40, 40] voxels (+/-80 mm).

---

Included Files

File	Description
rib_forest_pass1.joblib + feature_config_pass1.json	Pass-1 appearance random forest (120 trees, 228 features) and RSID configuration.
rib_forest_pass2.joblib + feature_config_img.json + feature_config_ctx.json	Pass-2 auto-context random forest (100 trees, 306 features) and appearance/context RSID configurations.
rib_atlas.joblib	24-rib statistical shape atlas (mean canonical curves, 5-D descriptor priors, inter-rib spacing priors).
benchmark_val12_fixed_gt.json, benchmark_val12_original_gt.json, benchmark_train47.json, benchmark_cv5.json	Full per-subject evaluation reports on the 12-scan test set, 47-scan training set, and 59-scan 5-fold cross-validation.
polar_stats_all59.json, gt_fix_log.json	Canonical frame verification statistics and ground-truth audit log.

---

Quick Start

from ribline import RibCageStatisticalAtlas, RibForest

atlas = RibCageStatisticalAtlas.load("rib_atlas.joblib")
rf_pass1 = RibForest.load("rib_forest_pass1.joblib")
rf_pass2 = RibForest.load("rib_forest_pass2.joblib")

---

Citation

• Paper: https://arxiv.org/abs/2610.16047 (PDF)

@article{wang2026hybrid,
  title={Hybrid Context Forests and Bilateral Viterbi Matching for {3D} Rib Centerline Tracing in {CT}},
  author={Wang, Quan},
  journal={arXiv preprint arXiv:2610.16047},
  year={2026},
  url={https://arxiv.org/abs/2610.16047}
}

@inproceedings{wu2012learning,
  title={A learning based deformable template matching method for automatic rib centerline extraction and labeling in {CT} images},
  author={Wu, Dijia and Liu, David and Pusk{\'a}s, Zsolt and Lu, Caiyun and Wimmer, Andreas and Tietjen, Christian and Soza, Grzegorz and Zhou, S Kevin},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={980--987},
  year={2012},
  organization={IEEE}
}

@inproceedings{wang2013semantic,
  title={Semantic context forests for learning-based knee cartilage segmentation in {3D} {MR} images},
  author={Wang, Quan and Wu, Dijia and Lu, Le and Liu, Meizhu and Boyer, Kim L and Zhou, Shaohua Kevin},
  booktitle={International MICCAI Workshop on Medical Computer Vision},
  pages={105--115},
  year={2013},
  organization={Springer}
}