Method

Memory-Augmented Segment Anything Model with Identity Consistency for MOTS [MASAM]
[Anonymous Submission]

Submitted on 7 Jul. 2025 10:06 by
[Anonymous Submission]

Running time:0.44 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
We propose a memory-augmented framework that adapts
SAM2 for performing identity-aware multi-object
tracking and segmentation without the need for fine-
tuning.
Parameters:
TBD
Latex Bibtex:

Detailed Results

From all 29 test sequences, our benchmark computes the commonly used tracking metrics (adapted for the segmentation case): CLEARMOT, MT/PT/ML, identity switches, and fragmentations [1,2]. The tables below show all of these metrics.


Benchmark sMOTSA MOTSA MOTSP MODSA MODSP
CAR 76.00 % 87.90 % 86.90 % 88.20 % 89.50 %
PEDESTRIAN 60.50 % 78.80 % 79.10 % 79.50 % 93.10 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 91.00 % 97.00 % 93.90 % 33463 1026 3297 9.20 % 49091 1029
PEDESTRIAN 87.50 % 91.60 % 89.50 % 18117 1660 2580 15.00 % 27230 640

Benchmark MT PT ML IDS FRAG
CAR 76.40 % 19.50 % 4.10 % 130 509
PEDESTRIAN 68.90 % 25.60 % 5.60 % 140 465

This table as LaTeX

[1] K. Bernardin, R. Stiefelhagen: Evaluating Multiple Object Tracking Performance: The CLEAR MOT Metrics. JIVP 2008.
[2] Y. Li, C. Huang, R. Nevatia: Learning to associate: HybridBoosted multi-target tracker for crowded scene. CVPR 2009.


eXTReMe Tracker