ECCV 2026 accepted papers on Architectures and efficiency
70 of the 2,834 papers accepted at ECCV 2026 (2.5%) are on Architectures and efficiency, compared with 4.0% at ECCV 2024. They fall into 7 topics.
See ECCV 2026 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest papers by meaning and how crowded the spot has become.
Topics and papers
Attention and transformers · 23 papers
- Pay Attention to Attention Distribution: A New Local Lipschitz Bound for Transformers (12 citations)
- OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer (8 citations)
- Vision-TTT: Efficient and Expressive Visual Representation Learning with Test-Time Training (4 citations)
- FILT3R: Latent State Adaptive Kalman Filter for Streaming 3D Reconstruction (2 citations)
- AVQ-Attention: Adaptive Vector-Quantized Attention (1 citation)
- Fast Spatial Memory with Scalable Elastic Test-Time Training (1 citation)
- Less Tokens, Better Forecasts: Sparse Residual Routing for Efficient Weather Prediction (1 citation)
- SAF3R: Dynamic Sparse Attention for Feed-Forward 3D Reconstruction Transformers (1 citation)
- Seeing Through Circuits: Faithful Mechanistic Interpretability for Vision Transformers (1 citation)
- When Token Compression Breaks: Structural Pruning vs. Token Reduction for Robust ViT Segmentation under High Compression (1 citation)
- CAR-MIL: Counterfactual Attention Regularization for Multiple Instance Learning
- Compact and Structurally Transparent Cervical Cytology with Geometry-Driven Features and Closed-Form Attention
and 11 more in the interactive map.
Pruning and efficient tuning · 19 papers
- SIMPLER: Efficient Foundation Model Adaptation via Similarity-Guided Layer Pruning for Earth Observation (2 citations)
- Low-Rank Ternary Adaptation for Fine-Tuning Transformers (1 citation)
- Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging (1 citation)
- AFFMAE: Scalable Vision Pre-Training for High-Resolution Microscopy Segmentation on Desktop Hardware
- Beyond Filter Pruning: Top-K Spatial Selection for Efficient Neural Networks
- CaRe: Critical Parameter Rectification for Efficient Visual Modeling
- Condensing Large-Scale Datasets Directly with Minimal Information Loss
- Curvature-Guided Mixing for MLLM Adaptation
- Denoised Variance-Based Pruning with Optimal Brain Bias Compensation
- Layering Virtual Try-On
- LoCA: Spatially-Aware Low-Rank Convolutional Adaptation of Vision Foundation Models
- MagicPrompt: Ultra-Lightweight Prompt Tuning for Video Generation
and 7 more in the interactive map.
RNNs and normalization · 11 papers
- SLAM-Former: Putting SLAM into One Transformer (6 citations)
- RT-DocLayout: Real-Time End-to-End Document Layout Analysis with Reading Order in the Wild (3 citations)
- FreeFlow: A Bias-free Hierarchical Transformer for Optical Flow Estimation
- Learn to Rank: Visual Attribution by Learning Importance Ranking
- Learning 1-Bit LiDAR-based Localization with Auxiliary Objective
- Robustness Emerges Early in Training Dynamics, but Is Not Preserved
- SafeSAE-VLA: Interpreting OpenVLA Progress Dynamics with Sparse Feature Analysis
- Spectral Gating via Damped Oscillations for Adaptive Implicit Neural Representations
- Weight-Space Mixture-of-Experts for Implicit Neural Representation Classification
- SPICE: Simple Polysemantic feature Interpretation via Clustering-based Explanations
- Weight Feedback Computes the Exact Jacobian Transpose in Modern Deep Networks
Mixture of experts · 8 papers
- TSEmbed: Unlocking Task Scaling in Universal Multimodal Embeddings (11 citations)
- Learning to Recover Task Experts from a Multi-Task Merged Model (1 citation)
- Training-Free Task Classification for Multi-Task Model Merging (1 citation)
- ExPLoRe: Expert Patch-Level Loss Routing for Multi-Objective Masked Image Modeling
- GeMoE: Gating Entropy is All You Need for Uncertainty-aware Adaptive Routing in MoE-based Large Vision-Language Models
- MixCompress: Mixture of Experts for Variable Rate Learned Image Compression
- MoE-ViE: Mixture of Experts Vision Encoder for Efficient Image and Video Understanding
- Preventing Expert Collapse in MoE-dVLMs via Modality-Wise Norm Alignment
Quantization · 7 papers
- BATQuant: Outlier-Resilient MXFP4 Quantization via Learnable Block-wise Optimization (6 citations)
- BLOB-Q: Boosting Low Bit ViT Quantization via Global Optimization on Model Distortion (1 citation)
- DiffPro: Joint Timestep and Layer-Wise Precision Optimization for Efficient Diffusion Inference (1 citation)
- Accurate Zero-shot Quantization via Hierarchical Teacher-Assistant Distillation
- Efficient Quantization-Aware Adaptation for Visual Foundation Models
- ORFC: Orthogonal Reparameterization for Low-Bitrate ViT Feature Coding
- Stabilizing Ultra-Low-Bit Quantization of Multimodal LLMs via Global Bit Allocation
