AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

SSAL: Synergizing between Self-Training and Adversarial Learning for Domain Adaptive Object Detection

Muhammad Akhtar Munir, Muhammad Haris Khan, M. Saquib Sarfraz, Mohsen Ali

NeurIPS 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Domain-adaptive object detection · 32 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
combine uncertainty-guided pseudo-labeling and uncertainty-guided tilingFCOS domain adaptation · Sim10K → CityscapesAP@0.5: 46.0 -> 51.8Improved
add uncertainty-guided pseudo-labeling (UGPL) (ablation)FCOS domain adaptation · Sim10K → CityscapesAP@0.5: 46.0 -> 49.5Improved
add uncertainty-guided tiling (UGT) (ablation)FCOS domain adaptation · Sim10K → CityscapesAP@0.5: 46.0 -> 50.0Improved
use confidence-based pseudo-label selection instead of uncertainty-guided selection (ablation)FCOS domain adaptation · Sim10K → CityscapesAP@0.5: 49.5 -> 43.2Hurt
use uncertainty-guided tiles instead of full images, random tiles, or certain tiles (ablation)UGPL adversarial alignment · Sim10K → CityscapesAP@0.5: 48.1 / 49.8 / 50.2 -> 51.8Improved
add R1 and R2 rounds with both UGPL and UGT after R0 (ablation)FCOS domain adaptation · Sim10K → CityscapesAP@0.5: 46.3 -> 51.8Improved
increase spatial dropout from 10% to 20% or 30% (ablation)SSAL · Sim10K → CityscapesAP@0.5: 51.8 -> 50.3; 51.8 -> 49.4Hurt
vary uncertainty threshold κ₁ from 0.5 to 0.4 or 0.6 (ablation)SSAL · Sim10K → CityscapesAP (mean): 28.9 -> 28.6; 28.9 -> 28.3Hurt
increase IoU threshold γ from 0.5 to 0.6 or 0.7 (ablation)SSAL · Sim10K → CityscapesAP (mean): 28.9 -> 28.3; 28.9 -> 27.5Hurt
adapt with SSAL rather than EPMFCOS domain adaptation · Sim10K → CityscapesAP@0.5: 49.0 -> 51.8Improved
adapt with SSAL rather than EPMFCOS domain adaptation · KITTI → CityscapesAP@0.5: 43.2 -> 45.6Improved
adapt with SSAL rather than the strongest reported baselineFCOS domain adaptation · Cityscapes → Foggy CityscapesmAP@0.5: 40.9 -> 39.6Hurt

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