Attention mechanisms, vision transformers, and the architectures replacing convolutions across vision tasks. We unpack how attention is used, misused, and reinvented in current research, from efficient attention branches to Mamba-style state space models.
Weak-Mamba-UNet: How CNN, ViT, and Visual Mamba Collaborate to Segment Medical Images from Scribbles | AI Trend Blend Medical AI & Weakly-Supervised Learning · arXiv:2402.10887 · University of Oxford / Mianyang Visual Engineering Center · 25 min read Teaching Three Different Brains to Agree — How Weak-Mamba-UNet Segments Hearts from Scribbles Ziyang Wang at Oxford […]
Mamba-3: Three Simple Ideas That Finally Fix What Transformers Get Wrong at Inference | AI Trend Blend AITrendBlend Machine Learning NLP & LLMs About Efficient AI · arXiv:2603.15569 · CMU & Princeton · March 2026 · 22 min read Mamba-3: Three Simple Ideas That Finally Fix What Transformers Get Wrong at Inference Time Researchers at
Mamba-3: Three Simple Ideas That Finally Fix What Transformers Get Wrong at Inference Read More »
GateMamba: Feature Gated Mixer in State Space Model for Point Cloud 3D Object Detection | AI Trend Blend AITrendBlend Machine Learning Computer Vision About Autonomous Driving AI · ISPRS Journal of Photogrammetry and Remote Sensing 236 (2026) 640–653 · 22 min read GateMamba: How Three Gated Mixers Taught a Mamba Network to Stop Ignoring Cyclists
GateMamba: Feature Gated Mixer in State Space Model for Point Cloud 3D Object Detection Read More »
The Moon’s Many Faces: A Single Unified Transformer for Multimodal Lunar Reconstruction | AI Trend Blend Planetary AI & 3D Reconstruction · ISPRS J. Photogramm. Remote Sens. 236 (2026) 363–379 · TU Dortmund University · 26 min read The Moon’s Many Faces: How One Transformer Learned to Speak All Four Languages of Lunar Science Simultaneously
The Moon’s Many Faces: A Single Unified Transformer for Multimodal Lunar Reconstruction Read More »
Fusion-Mamba: Hidden State Space Fusion for Cross-Modality Object Detection | AI Trend Blend AITrendBlend Machine Learning Computer Vision About Computer Vision · arXiv:2404.09146 · Beihang University · 21 min read Mamba Goes Multimodal: How Fusion-Mamba Built a Hidden State Space to End Modality Disparity Researchers at Beihang University asked what happens when you stop treating
Fusion-Mamba: Hidden State Space Fusion for Cross-Modality Object Detection Read More »
BGPANet: How Bi-Granular Progressive Attention Cracked the Skin Cancer Diagnosis Problem | AI Medical Research AIMedical Research Machine Learning Medical AI About Medical Image AI · Expert Systems With Applications 321 (2026) 132169 · 16 min read BGPANet: The Bi-Granular Attention Breakthrough That Finally Taught AI to Diagnose Skin Cancer Like a Dermatologist How a
BGPANet: How Bi-Granular Progressive Attention Cracked the Skin Cancer Diagnosis Problem Read More »
CFFormer: How Cross CNN-Transformer Attention Finally Solves the Blurry Ultrasound Problem | AI Trend Blend AITrendBlend Machine Learning Computer Vision Medical AI About Medical Image Segmentation · Expert Systems with Applications · 2025 · 24 min read CFFormer: How Cross CNN-Transformer Attention Finally Solves the Blurry Ultrasound Problem Researchers at University of Nottingham Ningbo built
Analysis by the aitrendblend editorial team · Medical image segmentation · Computational Visual Media, 2026 PraNet-V2 Dual Supervised Reverse Attention Polyp Segmentation Multi Organ CT Cardiac MRI Overview of the PraNet-V2 decoder. Three cascaded DSRA stages refine a coarse prediction using both a foreground head and an independently supervised background head. A flat polyp sitting
PraNet-V2 Fixes Medical Segmentation By Modeling Background Read More »
WEMoE: How a Mixture-of-Experts Approach Is Solving the Multi-Task Model Merging Problem | MedAI Research Deep Learning · TPAMI, 2026 · 18 min read The Static Model Merging Problem — and How WEMoE Learned to Adapt WEMoE introduces a dynamic mixture-of-experts approach to multi-task model merging, transforming how we combine fine-tuned neural networks by routing
WEMoE: How a Mixture-of-Experts Approach Is Solving the Multi-Task Model Merging Problem Read More »
AI for medical imaging and healthcare Vision foundation models CT and MRI segmentation Self supervised pretraining Analysis by the aitrendblend editorial team A radiation oncologist planning a course of treatment needs the kidneys, liver, spinal cord, and every nearby organ outlined precisely enough that the radiation beam avoids them by design rather than by luck.
MedDINOv3 Adapts A Vision Foundation Model For CT And MRI Segmentation Read More »