Method

SAM-STEP [SAM-STEP]
[Anonymous Submission]

Submitted on 24 Jun. 2026 18:22 by
[Anonymous Submission]

Running time:1 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
SAM-STEP is a hybrid and modular framework for the
STEP task that pairs a fine-tuned foundation
segmentation model with explicit, non-learned
association and refinement rules. SAM3's promptable
concept segmentation is adapted to a fixed dataset
taxonomy, temporally consistent and occlusion-robust
identities are derived through IoU-based propagation
matching and remaining segmentation gaps are closed
with superpixel-guided mask dilation.
Parameters:
\Gamma=\mathcal{C}_\text{void,crowd}
\sigma=0.35
\tau_f=0.65,w=5
\tau_{st}=0.6
g_{\max}=30
\tau_{mt}=0.8
\tau_\rho=0.3
\tau_d=0.6
\tau_r=0.55

Runtime is not measured.
Latex Bibtex:

Detailed Results

From all 29 test sequences, our benchmark computes the STQ segmentation and tracking metric (STQ, AQ, SQ (IoU)). The tables below show all of these metrics.


Benchmark STQ AQ SQ (IoU)
KITTI-STEP 67.94 % 71.39 % 64.67 %

This table as LaTeX



eXTReMe Tracker