SORT_TAO: Simple Online Realtime Tracker - Tuned on TAO train


Video not available.

Short name:

SORT_TAO

Description:

SORT (https://github.com/abewley/sort) applied on top of an LVIS+COCO-trained ResNet 50 Mask R-CNN. See paper for details: https://arxiv.org/abs/2005.10356.
The runtime is estimated at 2 days for running the detector on 2 4-gpu machines, and 14 hours for running SORT.

Reference:

A. Dave, T. Khurana, P. Tokmakov, C. Schmid, D. Ramanan. TAO: A Large-Scale Benchmark for Tracking Any Object. In The European Conference on Computer Vision (ECCV), 2020.

Last submitted:

July 10, 2020 (1 month ago)

Published:

July 10, 2020 at 17:44:11 CET

Submissions:

1

Project page / code:

Open source:

Yes

Hardware:

2.4GHZ, 56 cores

Runtime:

16.3 Hz

Benchmark performance:

Sequence TrackAP50 TrackAP75 TrackAP50_95 TrackAP50_area_small TrackAP50_area_medium TrackAP50_area_large
ECCV 2020 TAO Challenge10.2174.3834.8927.7208.22715.166

Detailed performance:

Sequence TrackAP50 TrackAP75 TrackAP50_95 TrackAP50_area_small TrackAP50_area_medium TrackAP50_area_large
TAO_test10.2174.3834.8927.7208.22715.166

Raw data: