CVPR 2023 accepted papers on Architectures and efficiency
117 of the 2,359 papers accepted at CVPR 2023 (5.0%) are on Architectures and efficiency, compared with 6.6% at CVPR 2022. They fall into 8 topics.
See CVPR 2023 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 · 41 papers
- BiFormer: Vision Transformer With Bi-Level Routing Attention (1,147 citations)
- EfficientViT: Memory Efficient Vision Transformer With Cascaded Group Attention (879 citations)
- Learning a Sparse Transformer Network for Effective Image Deraining (525 citations)
- Neighborhood Attention Transformer (474 citations)
- Lite-Mono: A Lightweight CNN and Transformer Architecture for Self-Supervised Monocular Depth Estimation (378 citations)
- Generalized Relation Modeling for Transformer Tracking (247 citations)
- Revealing the Dark Secrets of Masked Image Modeling (116 citations)
- R2Former: Unified Retrieval and Reranking Transformer for Place Recognition (111 citations)
- Comprehensive and Delicate: An Efficient Transformer for Image Restoration (107 citations)
- Slide-Transformer: Hierarchical Vision Transformer With Local Self-Attention (105 citations)
- FlexiViT: One Model for All Patch Sizes (91 citations)
- PHA: Patch-Wise High-Frequency Augmentation for Transformer-Based Person Re-Identification (90 citations)
and 29 more in the interactive map.
RNNs and normalization · 22 papers
- Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks (2,281 citations)
- Revisiting Reverse Distillation for Anomaly Detection (259 citations)
- FlowFormer++: Masked Cost Volume Autoencoding for Pretraining Optical Flow Estimation (113 citations)
- A Light Weight Model for Active Speaker Detection (56 citations)
- Hierarchical Neural Memory Network for Low Latency Event Processing (39 citations)
- PyPose: A Library for Robot Learning With Physics-Based Optimization (35 citations)
- You Need Multiple Exiting: Dynamic Early Exiting for Accelerating Unified Vision Language Model (34 citations)
- FFCV: Accelerating Training by Removing Data Bottlenecks (31 citations)
- Efficient On-Device Training via Gradient Filtering (22 citations)
- Towards a Smaller Student: Capacity Dynamic Distillation for Efficient Image Retrieval (22 citations)
- TINC: Tree-Structured Implicit Neural Compression (21 citations)
- Stitchable Neural Networks (20 citations)
and 10 more in the interactive map.
Pruning and efficient tuning · 19 papers
- DepGraph: Towards Any Structural Pruning (484 citations)
- TinyMIM: An Empirical Study of Distilling MIM Pre-Trained Models (47 citations)
- BlackVIP: Black-Box Visual Prompting for Robust Transfer Learning (43 citations)
- Visual Query Tuning: Towards Effective Usage of Intermediate Representations for Parameter and Memory Efficient Transfer Learning (40 citations)
- CLIPPING: Distilling CLIP-Based Models With a Student Base for Video-Language Retrieval (33 citations)
- Structured Sparsity Learning for Efficient Video Super-Resolution (33 citations)
- Master: Meta Style Transformer for Controllable Zero-Shot and Few-Shot Artistic Style Transfer (30 citations)
- Visual Atoms: Pre-Training Vision Transformers With Sinusoidal Waves (25 citations)
- Trainable Projected Gradient Method for Robust Fine-Tuning (19 citations)
- Adapting Shortcut With Normalizing Flow: An Efficient Tuning Framework for Visual Recognition (16 citations)
- Memory-Friendly Scalable Super-Resolution via Rewinding Lottery Ticket Hypothesis (12 citations)
- A-La-Carte Prompt Tuning (APT): Combining Distinct Data via Composable Prompting (10 citations)
and 7 more in the interactive map.
CNN design · 13 papers
- ConvNeXt V2: Co-Designing and Scaling ConvNets With Masked Autoencoders (1,589 citations)
- InternImage: Exploring Large-Scale Vision Foundation Models With Deformable Convolutions (982 citations)
- SCConv: Spatial and Channel Reconstruction Convolution for Feature Redundancy (762 citations)
- Learned Image Compression With Mixed Transformer-CNN Architectures (401 citations)
- CompletionFormer: Depth Completion With Convolutions and Vision Transformers (143 citations)
- FeatureBooster: Boosting Feature Descriptors With a Lightweight Neural Network (48 citations)
- DeepMAD: Mathematical Architecture Design for Deep Convolutional Neural Network (43 citations)
- DAA: A Delta Age AdaIN Operation for Age Estimation via Binary Code Transformer (23 citations)
- Learning Partial Correlation Based Deep Visual Representation for Image Classification (16 citations)
- SMPConv: Self-Moving Point Representations for Continuous Convolution (16 citations)
- Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations (9 citations)
- Long Range Pooling for 3D Large-Scale Scene Understanding (9 citations)
and 1 more in the interactive map.
Quantization · 11 papers
- PD-Quant: Post-Training Quantization Based on Prediction Difference Metric (75 citations)
- NoisyQuant: Noisy Bias-Enhanced Post-Training Activation Quantization for Vision Transformers (52 citations)
- LVQAC: Lattice Vector Quantization Coupled With Spatially Adaptive Companding for Efficient Learned Image Compression (25 citations)
- NIPQ: Noise Proxy-Based Integrated Pseudo-Quantization (20 citations)
- NVTC: Nonlinear Vector Transform Coding (19 citations)
- One-Shot Model for Mixed-Precision Quantization (19 citations)
- Solving Oscillation Problem in Post-Training Quantization Through a Theoretical Perspective (16 citations)
- Genie: Show Me the Data for Quantization (15 citations)
- CABM: Content-Aware Bit Mapping for Single Image Super-Resolution Network With Large Input (14 citations)
- Bit-Shrinking: Limiting Instantaneous Sharpness for Improving Post-Training Quantization (13 citations)
- ABCD: Arbitrary Bitwise Coefficient for De-Quantization (12 citations)
Neural architecture search · 8 papers
- Minimizing the Accumulated Trajectory Error To Improve Dataset Distillation (72 citations)
- Differentiable Architecture Search With Random Features (23 citations)
- EMT-NAS:Transferring Architectural Knowledge Between Tasks From Different Datasets (17 citations)
- MDL-NAS: A Joint Multi-Domain Learning Framework for Vision Transformer (17 citations)
- PA&DA: Jointly Sampling Path and Data for Consistent NAS (16 citations)
- HOTNAS: Hierarchical Optimal Transport for Neural Architecture Search (15 citations)
- Dynamic Neural Network for Multi-Task Learning Searching Across Diverse Network Topologies (7 citations)
- Auto-CARD: Efficient and Robust Codec Avatar Driving for Real-Time Mobile Telepresence (5 citations)