CVPR 2026 accepted papers on Architectures and efficiency
128 of the 4,042 papers accepted at CVPR 2026 (3.2%) are on Architectures and efficiency, compared with 5.0% at CVPR 2025. They fall into 8 topics.
See CVPR 2026 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
Attention and transformers · 39 papers
- BinaryAttention: One-Bit QK-Attention for Vision and Diffusion Transformers
- Can Natural Image Autoencoders Compactly Tokenize fMRI Volumes for Long-Range Dynamics Modeling?
- Can You Learn to See Without Images? Procedural Warm-Up for Vision Transformers
- Co-Me: Confidence Guided Token Merging for Visual Geometric Transformers
- Driving on Registers
- Efficient Equivariant Transformer for Self-Driving Agent Modeling
- Finding Distributed Object-Centric Properties in Self-Supervised Transformers
- FlashVGGT: Efficient and Scalable Visual Geometry Transformers with Compressed Descriptor Attention
- From Weights to Concepts: Data-Free Interpretability of CLIP via Singular Vector Decomposition
- Gated KalmaNet: A Fading Memory Layer through Test-time Ridge Regression
- HeSS: Head Sensitivity Score for Sparsity Redistribution in VGGT
- HTTM: Head-wise Temporal Token Merging for Faster VGGT
and 27 more in the interactive map.
Pruning and efficient tuning · 30 papers
- A Unified Framework for Knowledge Transfer in Bidirectional Model Scaling
- ACE-Merging: Data-Free Model Merging with Adaptive Covariance Estimation
- Basis-Oriented Low-rank Transfer for Few-Shot and Test-Time Adaptation
- Chain-of-Models Pre-Training: Rethinking Training Acceleration of Vision Foundation Models
- DC-Merge: Improving Model Merging with Directional Consistency
- Decompose, Mix, Adapt: A Unified Framework for Parameter-Efficient Neural Network Recombination and Compression
- Defending Unauthorized Model Merging via Dual-Stage Weight Protection
- ELiC: Efficient LiDAR Geometry Compression via Cross-Bit-depth Feature Propagation and Bag-of-Encoders
- FAAR: Efficient Frequency-Aware Multi-Task Fine-Tuning via Automatic Rank Selection
- FINE: Factorizing Knowledge for Initialization of Variable-sized Diffusion Models
- GFRRN: Explore the Gaps in Single Image Reflection Removal
- HiFICL: High-Fidelity In-Context Learning for Multimodal Tasks
and 18 more in the interactive map.
RNNs and normalization · 19 papers
- AdaBet: Gradient-free Layer Selection for Efficient Training of Deep Neural Networks
- Attribution-Guided Model Rectification of Unreliable Neural Network Behaviors
- Content-Aware Frequency Encoding for Implicit Neural Representations with Fourier-Chebyshev Features
- Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge
- Dataset Distillation by Influence Matching
- Efficient All-Pairs Correlation Volume Sampling for Optical Flow Estimation
- Hidden Monotonicity: Explaining Deep Neural Networks via their DC Decomposition
- Improving Sparse Autoencoder with Dynamic Attention
- Pip-Stereo: Progressive Iterations Pruner for Iterative Optimization based Stereo Matching
- Revisiting Model Stitching In the Foundation Model Era
- The Surprising Effectiveness of Noise Pretraining for Implicit Neural Representations
- Adaptive Depth Lightweight RGB-T Tracking with Holistic Token Routing
and 7 more in the interactive map.
Quantization · 13 papers
- AdaRadar: Rate Adaptive Spectral Compression for Radar-based Perception
- CAR-SAM: Cross-Attention Reconstruction for Post-Training Quantization of the Segment Anything Model
- Differentiable Vector Quantization for Rate-Distortion Optimization of Generative Image Compression
- QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models
- QVGGT: Post-Training Quantized Visual Geometry Grounded Transformer
- S2D: Selective Spectral Decay for Quantization-Friendly Conditioning of Neural Activations
- Spherical Leech Quantization for Visual Tokenization and Generation
- TWEO: Transformers Without Extreme Outliers Enables FP8 Training And Quantization For Dummies
- AdaSVD: Singular Value Decomposition with Adaptive Mechanisms for Large Multimodal Models
- Gradient Knows Best: Mixed-Precision Quantization via Gradient-Guided Bit Allocation for Super-Resolution
- Is Bin Generation Indispensable? A Bin-Generation-Free Dataset Quantization via Semantic Perspective
- LS-ViT: Least-Squares Hessian Based Block Reconstruction for Low-Bit Post-Training Quantization of Vision Transformers
and 1 more in the interactive map.
Mixture of experts · 10 papers
- ERMoE: Eigen-Reparameterized Mixture-of-Experts for Stable Routing and Interpretable Specialization (1 citation)
- Enhancing Mixture-of-Experts Specialization via Cluster-Aware Upcycling
- MoDES: Accelerating Mixture-of-Experts Multimodal Large Language Models via Dynamic Expert Skipping
- MoE-GRPO: Optimizing Mixture-of-Experts via Reinforcement Learning in Vision-Language Models
- NESTOR: A Nested MOE-based Neural Operator for Large-Scale PDE Pre-Training
- TAG-MoE: Task-Aware Gating for Unified Generative Mixture-of-Experts
- Teacher-Guided Routing for Sparse Vision Mixture-of-Experts
- DuetMerging: Synergizing Dynamic and Static Strategies for Mitigating Task Interference in Model Merging
- OneSparse: A Unified Framework for Sparse Activation Layers in Vision Models
- Soft Modality-Guided Expert Specialization in MoE-VLMs
Spiking networks · 7 papers
- Rethinking SNN Online Training and Deployment: Gradient-Coherent Learning via Hybrid-Driven LIF Model (1 citation)
- Reconstructing Spiking Neural Networks Using a Single Neuron with Autapses
- SDTrack: A Baseline for Event-based Tracking via Spiking Neural Networks
- Stable Spike: Dual Consistency Optimization via Bitwise AND Operations for Spiking Neural Networks
- Robust Spiking Neural Networks by Temporal Mutual Information
- Temporal Interaction in Spiking Transformers with Multi-Delay Mixer
- Temporal Representation Enhancement (TRE): Learning to Forget Dominant Patterns for Enhanced Temporal Spiking Features
Neural architecture search · 5 papers
- HyperNAS: Enhancing Architecture Representation for NAS Predictor via Hypernetwork
- InTrain: Intrinsic Trainability for Zero-Cost Neural Architecture Search
- TAS-LoRA: Transformer Architecture Search with Mixture-of-LoRA Experts
- Progressive Neural Architecture Generation
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation
CNN design · 5 papers
- PFGNet: A Fully Convolutional Frequency-Guided Peripheral Gating Network for Efficient Spatiotemporal Predictive Learning
- Unified Spherical Frontend: Learning Rotation-Equivariant Representations of Spherical Images from Any Camera
- Consensus vs. Controversy: Mapping the Decision Space Where Architectures Diverge
- FBTA: Enabling Single-GPU End-to-End Gigapixel WSI Classification with Feature Bridging and Translation Alignment
- QuCNet: Quantum Deep Learning Driven Multi-Circuit Network for Remote Sensing Image Classification