AI Paper Atlas

Research map / Detection and segmentation / CNN features

M4-SAM: Multi-Modal Mixture-of-Experts with Memory-Augmented SAM for RGB-D Video Salient Object Detection

Jiyuan Liu, Jia Lin, Xiaofei Zhou, Runmin Cong, Deyang Liu, Zhi Liu

CVPR 2026 · Detection and segmentation › CNN features · approach: RGB-D salient object detection

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 actual depth instead of RGB copy as pseudo depth (ablation)M4-SAM · RDVSE-measure: 0.858 -> 0.927Improved
use actual depth instead of all-zero depth (ablation)M4-SAM · RDVSE-measure: 0.895 -> 0.927Improved
fine-tune encoder with Modality-Aware MoE-LoRA vs frozen (ablation)M4-SAM · RDVSE-measure: 0.839 -> 0.927Improved
Modality-Aware MoE-LoRA vs vanilla LoRA (ablation)M4-SAM · RDVSE-measure: 0.904 -> 0.927Improved
Modality-Aware MoE-LoRA vs LoRA training memory (ablation)M4-SAM · RDVSGPU memory (GB): 14.25 -> 11.62Improved
Conv-LoRA instead of vanilla LoRA (ablation)SAM2 encoder · RDVSE-measure: 0.904 -> 0.900No clear effect
MoE top-K 2 instead of 1 (ablation)M4-SAM · RDVSE-measure: 0.903 -> 0.927Improved
MoE top-K 2 instead of 3 (ablation)M4-SAM · RDVSE-measure: 0.914 -> 0.927Improved
add memory bank to baseline (ablation)M4-SAM · RDVSE-measure: 0.908 -> 0.910No clear effect
add Gated Multi-Level Feature Fusion to memory model (ablation)M4-SAM · RDVSE-measure: 0.910 -> 0.927Improved
use mid-level decoder feature X_D^2 instead of X_D^1 (ablation)M4-SAM · RDVSE-measure: 0.918 -> 0.927Improved
use both X_D^1 and X_D^2 (ablation)M4-SAM · RDVSE-measure: 0.927 -> 0.909Hurt

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