TrctrD15: deepMOTTracktor15

TUD-Crossing


Benchmark:

Short name:

TrctrD15

Detector:

Public

Description:

The recent trend in vision-based multi-object tracking (MOT) is heading towards leveraging the representational power of deep learning to jointly learn to detect and track objects. However, existing methods train only certain sub-modules using loss functions that often do not correlate with established tracking evaluation measures such as Multi-Object Tracking Accuracy (MOTA) and Precision (MOTP). As these measures are not differentiable, the choice of appropriate loss functions for end-to-end training of multi-object tracking methods is still an open research problem. In this paper, we bridge this gap by proposing a differentiable proxy of MOTA and MOTP, which we combine in a loss function suitable for end-to-end training of deep multi-object trackers. As a key ingredient, we propose a Deep Hungarian Net (DHN) module that approximates the Hungarian matching algorithm. DHN allows to estimate the correspondence between object tracks and ground truth objects to compute differentiable proxies of MOTA and MOTP, which are in turn used to optimize deep trackers directly. We experimentally demonstrate that the proposed differentiable framework improves the performance of existing multi-object trackers, and we establish a new state-of-the-art on the MOTChallenge benchmark.

Reference:

Y. Xu, A. Osep, Y. Ban, R. Horaud, L. Leal-Taixe, X. Alameda-Pineda. How To Train Your Deep Multi-Object Tracker. In , 2019.

Last submitted:

February 05, 2020 (8 months ago)

Published:

April 16, 2020 at 14:18:39 CET

Submissions:

3

Open source:

Yes

Hardware:

GTX-TITAN XP

Runtime:

1.6 Hz

Benchmark performance:

Sequence MOTA IDF1 MOTP MT ML FP FN Recall Precision FAF ID Sw. Frag
2D MOT 201544.146.075.3124 (17.2)192 (26.6)6,08526,91756.285.01.11,347 (24.0)1,868 (33.2)

Detailed performance:

Sequence MOTA IDF1 MOTP MT ML FP FN Recall Precision FAF ID Sw. Frag
ADL-Rundle-134.551.473.51032,3603,68760.470.44.753162
ADL-Rundle-345.544.881.3877434,72253.688.01.27693
AVG-TownCentre34.335.269.633447233,31853.684.11.6657668
ETH-Crossing43.958.483.12102053346.995.90.11019
ETH-Jelmoli57.566.379.4161230875270.485.30.71947
ETH-Linthescher49.155.680.430991584,33751.496.70.15193
KITTI-1649.747.573.51111369859.089.90.54479
KITTI-1949.558.371.59105392,08960.985.80.571138
PETS09-S2L247.129.371.5314804,31655.291.71.1307492
TUD-Crossing77.547.476.1801121480.698.80.12323
Venice-136.139.074.1456302,25150.778.61.43654

Raw data: