CVPR 2025 accepted papers on Architectures and efficiency
144 of the 2,871 papers accepted at CVPR 2025 (5.0%) are on Architectures and efficiency, compared with 4.4% at CVPR 2024. They fall into 8 topics.
See CVPR 2025 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 · 44 papers
- MambaOut: Do We Really Need Mamba for Vision? (188 citations)
- Transformers without Normalization (157 citations)
- MobileMamba: Lightweight Multi-Receptive Visual Mamba Network (91 citations)
- LSNet: See Large, Focus Small (90 citations)
- EfficientViM: Efficient Vision Mamba with Hidden State Mixer based State Space Duality (69 citations)
- Similarity-Guided Layer-Adaptive Vision Transformer for UAV Tracking (56 citations)
- Your ViT is Secretly an Image Segmentation Model (37 citations)
- 2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification (36 citations)
- Learning Occlusion-Robust Vision Transformers for Real-Time UAV Tracking (36 citations)
- BHViT: Binarized Hybrid Vision Transformer (32 citations)
- Trajectory Mamba: Efficient Attention-Mamba Forecasting Model Based on Selective SSM (27 citations)
- Building Vision Models upon Heat Conduction (25 citations)
and 32 more in the interactive map.
RNNs and normalization · 30 papers
- SuperLightNet: Lightweight Parameter Aggregation Network for Multimodal Brain Tumor Segmentation (22 citations)
- Dataset Distillation with Neural Characteristic Function: A Minmax Perspective (13 citations)
- MultiMorph: On-demand Atlas Construction (8 citations)
- ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model (6 citations)
- DeepCompress-ViT: Rethinking Model Compression to Enhance Efficiency of Vision Transformers at the Edge (5 citations)
- STINR: Deciphering Spatial Transcriptomics via Implicit Neural Representation (5 citations)
- Tuning the Frequencies: Robust Training for Sinusoidal Neural Networks (5 citations)
- Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging (4 citations)
- Driving by the Rules: A Benchmark for Integrating Traffic Sign Regulations into Vectorized HD Map (3 citations)
- End-to-End Implicit Neural Representations for Classification (3 citations)
- Comprehensive Information Bottleneck for Unveiling Universal Attribution to Interpret Vision Transformers (2 citations)
- Efficient Depth Estimation for Unstable Stereo Camera Systems on AR Glasses (2 citations)
and 18 more in the interactive map.
Pruning and efficient tuning · 28 papers
- Task Singular Vectors: Reducing Task Interference in Model Merging (12 citations)
- Lessons and Insights from a Unifying Study of Parameter-Efficient Fine-Tuning (PEFT) in Visual Recognition (9 citations)
- TinyFusion: Diffusion Transformers Learned Shallow (8 citations)
- Meta-Learning Hyperparameters for Parameter Efficient Fine-Tuning (7 citations)
- Coeff-Tuning: A Graph Filter Subspace View for Tuning Attention-Based Large Models (6 citations)
- ATP: Adaptive Threshold Pruning for Efficient Data Encoding in Quantum Neural Networks (5 citations)
- Sensitivity-Aware Efficient Fine-Tuning via Compact Dynamic-Rank Adaptation (5 citations)
- Temporal Action Detection Model Compression by Progressive Block Drop (5 citations)
- Faster Parameter-Efficient Tuning with Token Redundancy Reduction (4 citations)
- Flexible Group Count Enables Hassle-Free Structured Pruning (4 citations)
- IterIS: Iterative Inference-Solving Alignment for LoRA Merging (4 citations)
- Less is More: Efficient Model Merging with Binary Task Switch (4 citations)
and 16 more in the interactive map.
CNN design · 14 papers
- Frequency Dynamic Convolution for Dense Image Prediction (92 citations)
- Adaptive Rectangular Convolution for Remote Sensing Pansharpening (50 citations)
- A General Adaptive Dual-level Weighting Mechanism for Remote Sensing Pansharpening (24 citations)
- ShiftwiseConv: Small Convolutional Kernel with Large Kernel Effect (22 citations)
- Are Spatial-Temporal Graph Convolution Networks for Human Action Recognition Over-Parameterized? (13 citations)
- EffiDec3D: An Optimized Decoder for High-Performance and Efficient 3D Medical Image Segmentation (9 citations)
- A Selective Re-learning Mechanism for Hyperspectral Fusion Imaging (8 citations)
- Rotation-Equivariant Self-Supervised Method in Image Denoising (5 citations)
- Frequency-Biased Synergistic Design for Image Compression and Compensation (4 citations)
- DnLUT: Ultra-Efficient Color Image Denoising via Channel-Aware Lookup Tables (3 citations)
- No Pains, More Gains: Recycling Sub-Salient Patches for Efficient High-Resolution Image Recognition (2 citations)
- Parameterized Blur Kernel Prior Learning for Local Motion Deblurring (2 citations)
and 2 more in the interactive map.
Quantization · 12 papers
- Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression (21 citations)
- APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers (6 citations)
- BlueLM-V-3B: Algorithm and System Co-Design for Multimodal Large Language Models on Mobile Devices (6 citations)
- Quantization without Tears (6 citations)
- FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation (5 citations)
- QuartDepth: Post-Training Quantization for Real-Time Depth Estimation on the Edge (5 citations)
- Enhancing Diversity for Data-free Quantization (4 citations)
- CASP: Compression of Large Multimodal Models Based on Attention Sparsity (2 citations)
- Pioneering 4-Bit FP Quantization for Diffusion Models: Mixup-Sign Quantization and Timestep-Aware Fine-Tuning (1 citation)
- Towards Lossless Implicit Neural Representation via Bit Plane Decomposition (1 citation)
- Data-Free Group-Wise Fully Quantized Winograd Convolution via Learnable Scales
- HOT: Hadamard-based Optimized Training
Spiking networks · 8 papers
- Spiking Transformer with Spatial-Temporal Attention (25 citations)
- STAA-SNN: Spatial-Temporal Attention Aggregator for Spiking Neural Networks (13 citations)
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (12 citations)
- Rethinking Spiking Self-Attention Mechanism: Implementing a-XNOR Similarity Calculation in Spiking Transformers (6 citations)
- Temporal Separation with Entropy Regularization for Knowledge Distillation in Spiking Neural Networks (6 citations)
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (4 citations)
- Efficient ANN-Guided Distillation: Aligning Rate-based Features of Spiking Neural Networks through Hybrid Block-wise Replacement (3 citations)
- VISTREAM: Improving Computation Efficiency of Visual Streaming Perception via Law-of-Charge-Conservation Inspired Spiking Neural Network (1 citation)
Mixture of experts · 5 papers
- Complexity Experts are Task-Discriminative Learners for Any Image Restoration (51 citations)
- SPMTrack: Spatio-Temporal Parameter-Efficient Fine-Tuning with Mixture of Experts for Scalable Visual Tracking (16 citations)
- DeRS: Towards Extremely Efficient Upcycled Mixture-of-Experts Models (3 citations)
- How to Merge Your Multimodal Models Over Time? (3 citations)
- Efficient Data Driven Mixture-of-Expert Extraction from Trained Networks (1 citation)
Neural architecture search · 3 papers
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (4 citations)
- Training-free Neural Architecture Search through Variance of Knowledge of Deep Network Weights (2 citations)
- L-SWAG: Layer-Sample Wise Activation with Gradients Information for Zero-Shot NAS on Vision Transformers (1 citation)