AI Paper Atlas

Research map / Detection and segmentation / Tracking

XS-VID: A Large-Scale Benchmark for Small Object Detection and Tracking in Videos

Jiahao Guo, Ziyang Xu, Lianjun Wu, Fei Gao, Wenyu Liu, Xinggang Wang

TPAMI 2026 · Detection and segmentation › Tracking · approach: Unified transformer tracking · 3 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 small-object techniques (upsampling, backbone, loss) (ablation)YOLOv8 · XS-VID testAP: 22.7 -> 24.5Improved
add Flow Mask Attention without mask supervision (ablation)YOLOv8 + small-object techniques · XS-VID testAP: 24.5 -> 24.3Hurt
add SAM2 mask auxiliary loss (ablation)YOLOFT with FMA · XS-VID testAP: 24.3 -> 25.2Improved
add video data augmentation (ablation)YOLOFT · XS-VID testAP: 25.2 -> 25.8Improved
use 3D conv instead of concatenation for temporal fusion (ablation)FMA temporal fusion · XS-VID testAP: 23.3 -> 24.7Improved
use ConvLSTM instead of 3D conv for temporal fusion (ablation)FMA temporal fusion · XS-VID testAP: 24.7 -> 23.5Hurt
add current-frame mask supervision (ablation)YOLOFT · XS-VID testAP: 23.2 -> 24.1Improved
add unweighted historical mask supervision (ablation)YOLOFT with current mask · XS-VID testAP: 24.1 -> 23.5Hurt
add half-weight historical mask supervision (ablation)YOLOFT with current mask · XS-VID testAP: 24.1 -> 24.6Improved
replace bilinear with ESPCN upsampling (ablation)detector · XS-VID testAP^s: 12.4 -> 13.7Improved
replace GIoU with NWD loss (ablation)detector · XS-VID testAP^s: 12.0 -> 12.9Improved
use 0.4*CIoU + 0.6*SAFit loss instead of GIoU (ablation)detector · XS-VID testAP^s: 12.0 -> 14.1Improved

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