UnsupTrack: Simple Unsupervised Multi-Object Tracking


Video not available.

Rendering of new sequences is currently deactivated due to heavy load.

Rendering of new sequences is currently deactivated due to heavy load.

Rendering of new sequences is currently deactivated due to heavy load.

Rendering of new sequences is currently deactivated due to heavy load.

Rendering of new sequences is currently deactivated due to heavy load.

Rendering of new sequences is currently deactivated due to heavy load.

Rendering of new sequences is currently deactivated due to heavy load.

Benchmark:

MOT17 | MOT16 | MOT20 |

Short name:

UnsupTrack

Detector:

Public

Description:

Multi-object tracking has seen a lot of progress recently, albeit with substantial annotation costs for developing better and larger labeled datasets. In this work, we remove the need for annotated datasets by proposing an unsupervised re-identification network, thus sidestepping the labeling costs entirely, required for training. Given unlabeled videos, our proposed method (SimpleReID) first generates tracking labels using SORT and trains a ReID network to predict the generated labels using crossentropy loss. We demonstrate that SimpleReID performs substantially better than simpler alternatives, and we recover the full performance of its supervised counterpart consistently across diverse tracking frameworks. The observations are unusual because unsupervised ReID is not expected to excel in crowded scenarios with occlusions, and drastic viewpoint changes. By incorporating our unsupervised SimpleReID with CenterTrack trained on augmented still images, we establish a new state-of-the-art performance on popular datasets like MOT16/17 without using tracking supervision, beating current best (CenterTrack) by 0.2-0.3 MOTA and 4.4-4.8 IDF1 scores. We further provide evidence for limited scope for improvement in IDF1 scores beyond our unsupervised ReID in the studied settings. Our investigation suggests reconsideration towards more sophisticated, supervised, end-to-end trackers by showing promise in simpler unsupervised alternatives.

Reference:

S. Karthik, A. Prabhu, V. Gandhi. Simple Unsupervised Multi-Object Tracking. In Arxiv, 2020.

Last submitted:

May 18, 2020 (4 years ago)

Published:

May 18, 2020 at 16:09:12 CET

Submissions:

1

Project page / code:

n/a

Open source:

No

Hardware:

GTX 1080Ti

Runtime:

1.9 Hz

Benchmark performance:

Sequence MOTA IDF1 HOTA MT ML FP FN Rcll Prcn AssA DetA AssRe AssPr DetRe DetPr LocA FAF ID Sw. Frag
MOT1662.458.547.0205 (27.0)242 (31.9)5,90961,98166.095.344.849.853.864.153.076.581.21.0588 (8.9)1,361 (20.6)

Detailed performance:

Sequence MOTA IDF1 HOTA MT ML FP FN Rcll Prcn AssA DetA AssRe AssPr DetRe DetPr LocA FAF ID Sw. Frag
MOT16-0141.548.239.3785293,19050.185.843.136.347.268.140.068.579.21.22443
MOT16-0378.768.654.790111,54920,55780.398.249.161.356.867.464.578.881.51.0178471
MOT16-0652.741.136.257747154,64959.790.630.244.056.839.447.872.680.40.695205
MOT16-0745.345.635.01291,5687,25955.585.332.438.739.156.543.466.779.13.1105231
MOT16-0832.836.231.9102360210,55336.991.134.130.241.062.831.778.284.31.089147
MOT16-1248.446.040.717392873,96952.293.840.940.660.952.743.377.984.00.32448
MOT16-1432.240.329.3127865911,80436.191.033.226.137.660.927.569.377.80.973216

Raw data: