AI Paper Atlas

Research map / Detection and segmentation / Tracking

QDTrack: Quasi-Dense Similarity Learning for Appearance-Only Multiple Object Tracking

Tobias Fischer, T. Huang, Jiangmiao Pang, Linlu Qiu, Haofeng Chen, Trevor Darrell, Fisher Yu

TPAMI 2023 · Detection and segmentation › Tracking · approach: Foundation-model and correlation-filter tracking · 182 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add quasi-dense matching (many negatives, one positive) (ablation)QDTrack · BDD100K val (ablation, Faster R-CNN R50)IDF1: 63.0 -> 66.8Improved
add multi-positive contrastive loss (ablation)QDTrack · BDD100K val (ablation, Faster R-CNN R50)IDF1: 66.8 -> 67.8Improved
replace cosine similarity with bi-softmax (ablation)QDTrack · BDD100K val (ablation, Faster R-CNN R50)IDF1: 67.8 -> 70.0Improved
add duplicate removal and backdrops (ablation)QDTrack · BDD100K val (ablation, Faster R-CNN R50)IDF1: 70.0 -> 71.5Improved
add all quasi-dense training and inference components (ablation)QDTrack · BDD100K val (ablation, Faster R-CNN R50)IDF1: 63.0 -> 71.5Improved
add IoU cue to appearance association (ablation)QDTrack · BDD100K valmMOTA: 36.6 -> 36.3No clear effect
add distance thresholding (ablation)QDTrack · MOT17 val (half, ablation)IDF1: 73.6 -> 74.5Improved
add tracklet merging (ablation)QDTrack · MOT17 val (half, ablation)IDF1: 74.5 -> 76.0Improved
add linear interpolation (ablation)QDTrack · MOT17 val (half, ablation)MOTA: 76.0 -> 76.8Improved
add stronger data augmentation to supervised video training (ablation)QDTrack · BDD100K valmMOTA: +1.2 points in mMOTA, +2.1 points in mIDF1Improved
train 2x schedule instead of 1x (ablation)QDTrack (Faster R-CNN R50-FPN) · BDD100K valMOTA: 65.7 -> 65.6No clear effect
use ResNet-101 instead of ResNet-50 (ablation)QDTrack (Faster R-CNN) · BDD100K valMOTA: 65.7 -> 66.2Improved

Later papers that compared with this method

6 later accepted papers report QDTrack as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy