AI Paper Atlas

Research map / Detection and segmentation / Weakly supervised detection and segmentation

SOOD++: Leveraging Unlabeled Data to Boost Oriented Object Detection

Dingkang Liang, Wei Hua, Chunsheng Shi, Zhikang Zou, Xiaoqing Ye, Xiang Bai

TPAMI 2025 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Pseudo-labeling for semi-supervised detection · 14 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
use SIDS sampling instead of Dense Teacher sampling (ablation)Dense pseudo-labeling · DOTA-V2.0 val 10% labeledmAP: 42.73 -> 44.32Improved
add GAW loss on top of SIDS (ablation)SOOD++ · DOTA-V2.0 val 10% labeledmAP: 44.32 -> 47.65Improved
add NGC on top of SIDS (ablation)SOOD++ · DOTA-V2.0 val 10% labeledmAP: 44.32 -> 47.82Improved
combine GAW and NGC with SIDS (ablation)SOOD++ · DOTA-V2.0 val 10% labeledmAP: 44.32 -> 49.14Improved
replace SOOD sampling with SIDS (ablation)SOOD · DOTA-V1.5 val 10% labeledmAP: 48.63 -> 49.44Improved
replace RAW with GAW (ablation)SOOD · DOTA-V1.5 val 10% labeledmAP: 48.63 -> 49.38Improved
replace GC with NGC (ablation)SOOD · DOTA-V1.5 val 10% labeledmAP: 48.63 -> 49.61Improved
use sample ratio 0.25 instead of 1.0 (ablation)SIDS · DOTA-V1.5 val 10% labeledmAP: 49.11 -> 50.48Improved
use IoU instead of classification score for background filtering (ablation)SIDS · DOTA-V1.5 val 10% labeledmAP: 49.41 -> 50.48Improved
vary IoU threshold from 0.1 to 0.3 (ablation)SIDS · DOTA-V1.5 val 10% labeledmAP: 50.48 -> 50.17No clear effect
add GPT-4.1 pseudo-labels to supervised baseline instead of SOOD++MLLM pseudo-labeling · DOTA-V2.0 val 10% labeledmAP: 38.78 -> 41.07Mixed
add unlabeled data with Unbiased Teacher in fully labeled settingUnbiased Teacher · DOTA-V2.0 val 100% labeledmAP: 61.04 -> 60.65Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 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