Re-identification for tracking using Faster RCNN + reID net

MOT17-01-DPM MOT17-03-DPM MOT17-06-DPM MOT17-07-DPM MOT17-08-DPM MOT17-12-DPM MOT17-14-DPM MOT17-01-FRCNN MOT17-03-FRCNN MOT17-06-FRCNN

Short name:

reID2track

Benchmark:

Description:

The algorithm is an re-idetification task that uses the faster R-CNN. The improvements are the increase in the resolution of the network and scaling of anchors. The network learns features for each object and these features using the faster RCNN algorithm and on top of that an identification is applied that creates features for each unique object for the purpose of re-identification.
The detection features from the conv-net are used for tracking. The algorithm uses cosine similarity for feature association. For tracking, simply the IDs the assigned based on the similarity threshold.

Hardware:

Intel(R) Xeon(R) CPU E5-2690v3 2.6GHz, 256GB RAM, Nvidia Quadro P6000

Detector:

Public

Processing:

Online

Last submitted:

May 07, 2018 (7 months ago)

Published:

May 07, 2018 at 11:41:19 CET

Submissions:

2

Open source:

No

Project page / code:

n/a

Reference:

Anonymous submission

Benchmark performance:

MOTAMOTPFAFMTMLFPFNID Sw.FragSpecificationsDetector
44.676.91.315.8 % 39.7 % 22,451284,2136,13413,786Intel(R) Xeon(R) CPU E5-2690v3 2.6GHz, 256GB RAM, Nvidia Quadro P6000Public
IDF1ID PrecisionID Recall
39.957.230.7

Detailed performance:

Sequence MOTA IDF1 MOTP FAF GT MT ML FP FN ID Sw Frag
MOT17-01-DPM20.426.773.20.2244.2 % 62.5 % 1054,99037177
MOT17-03-DPM37.526.276.12.41489.5 % 31.8 % 3,58359,6662,1293,184
MOT17-06-DPM38.635.574.40.32228.1 % 54.5 % 3166,819100359
MOT17-07-DPM26.930.874.41.3606.7 % 58.3 % 64711,519179588
MOT17-08-DPM20.525.279.61.2767.9 % 55.3 % 74215,931131286
MOT17-12-DPM35.343.977.20.5918.8 % 52.7 % 4065,16638148
MOT17-14-DPM14.419.275.20.31642.4 % 78.0 % 20515,518100457
MOT17-01-FRCNN27.341.576.53.02425.0 % 33.3 % 1,3673,2923376
MOT17-03-FRCNN56.847.978.01.114824.3 % 18.9 % 1,57643,2983761,071
MOT17-06-FRCNN54.643.878.60.422222.1 % 25.7 % 4454,769141345
MOT17-07-FRCNN32.837.474.82.76010.0 % 25.0 % 1,3599,740251564
MOT17-08-FRCNN22.231.580.11.2769.2 % 52.6 % 76615,58375185
MOT17-12-FRCNN36.649.278.30.79113.2 % 48.4 % 6014,86334130
MOT17-14-FRCNN18.430.671.33.61646.7 % 46.3 % 2,69311,998398796
MOT17-01-SDP38.540.873.62.52429.2 % 16.7 % 1,1052,725134284
MOT17-03-SDP72.152.677.60.814845.3 % 10.1 % 1,27227,0608302,396
MOT17-06-SDP56.746.076.50.522229.7 % 29.3 % 6224,309170332
MOT17-07-SDP44.745.575.52.16018.3 % 26.7 % 1,0278,043267694
MOT17-08-SDP28.529.778.21.17613.2 % 47.4 % 66414,211231517
MOT17-12-SDP40.855.178.40.99120.9 % 40.7 % 8054,27547198
MOT17-14-SDP29.632.572.22.91646.1 % 35.4 % 2,14510,438433999

Raw data:

n/a


reID2track