Method

Feature-Fused 3 Frame [FF3F]
[Anonymous Submission]

Submitted on 29 Jul. 2026 23:02 by
[Anonymous Submission]

Running time:.009 s
Environment:>8 cores @ >3.5 Ghz (Python)

Method Description:
FF3F-KITTI is an online modular multi-object
tracking system that processes RGB images and
Velodyne point clouds through separate specialized
branches. Confidence-aware fusion combines the
useful detection and geometric information, while
ego-motion compensation and a lightweight
AssociationNet connect detections across frames.
Class-specific routing enables or disables
branches according to their usefulness for Car and
Pedestrian tracking.
Parameters:
Car RGB detector: 4 epochs, AdamW optimizer,
initial learning rate = 0.0002, weight decay =
0.0005, batch size = 6, warm-up = 1 epoch, and
input resolution = 960 pixels. Each of the five
FusionNet heads was trained for 5 epochs using
AdamW, with learning rate = 0.001, weight decay =
0.0001, and batch size = 1024. AssociationNet was
trained for 10 epochs using AdamW, with learning
rate = 0.0005, weight decay = 0.0001, and matrix
batch size = 16.
Latex Bibtex:

Detailed Results

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


Benchmark MOTA MOTP MODA MODP
CAR 82.40 % 84.75 % 82.82 % 87.86 %
PEDESTRIAN 37.69 % 71.05 % 40.68 % 90.89 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 91.70 % 93.14 % 92.42 % 36014 2651 3259 23.83 % 42683 1796
PEDESTRIAN 66.01 % 72.70 % 69.19 % 15422 5790 7942 52.05 % 25526 1342

Benchmark MT PT ML IDS FRAG
CAR 73.08 % 22.92 % 4.00 % 143 385
PEDESTRIAN 34.02 % 49.83 % 16.15 % 692 1549

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