AI Paper Atlas

Research map / Detection and segmentation / CNN features

YOLO-ULM: Ultra-Lightweight Models for Real-Time Object Detection

Shasha Han, Chong Li, Xinning Wang, Xuebo Li

CVPR 2026 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors

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 D3C2f feature aggregation (ablation)YOLO-ULM-N · COCO valAP50:95: 40.7 -> 40.9Improved
add D3C2f (ablation)YOLO-ULM-M · COCO valAP50:95: 52.6 -> 52.6No clear effect
add RepDown downsampling (ablation)YOLO-ULM-N · COCO valAP50:95 / params: 40.9 -> 41.0; params 2.6M -> 2.1MImproved
add RepDown downsampling (ablation)YOLO-ULM-M · COCO valAP50:95 / params: 52.6 -> 52.5; params 20.2M -> 16.0MMixed
add FoCIoU loss (ablation)YOLO-ULM-N · COCO valAP50:95: 41.0 -> 41.2Improved
add FoCIoU loss (ablation)YOLO-ULM-M · COCO valAP50:95: 52.5 -> 52.8Improved
use FoCIoU instead of Focaler-IoU (ablation)YOLO-ULM-N · COCO valAP50:95: 40.9 -> 41.2Improved
use RepDown instead of ADown (ablation)YOLO-ULM-N · COCO valAP50:95: 40.7 -> 41.2Improved
use 7x7 kernel in RepDown instead of 5x5 or 9x9 (ablation)YOLO-ULM-N · COCO valAP50:95: 41.3 / 41.2 -> 41.6Improved
use padding in RepDown with 600 epochs (ablation)YOLO-ULM-N · COCO valAP50:95: 41.6 -> 41.2Hurt
use padding in RepDown with 600 epochs (ablation)YOLO-ULM-S · COCO valAP50:95: 47.8 -> 48.1Improved
use grouped convolution in downsampling (ablation)YOLO-ULM-L/X · COCO valparams / AP50:95: L 54.1 -> 54.1; X 55.7 -> 55.6Improved

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