CVPR 2022 accepted papers on Architectures and efficiency
137 of the 2,077 papers accepted at CVPR 2022 (6.6%) are on Architectures and efficiency, compared with 6.0% at CVPR 2021. They fall into 7 topics.
See CVPR 2022 in the interactive map
Submitting to CVPR? Checklist →
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 · 55 papers
- Masked Autoencoders Are Scalable Vision Learners (7,924 citations)
- Uformer: A General U-Shaped Transformer for Image Restoration (2,236 citations)
- CSWin Transformer: A General Vision Transformer Backbone With Cross-Shaped Windows (1,316 citations)
- MetaFormer Is Actually What You Need for Vision (1,239 citations)
- SimMIM: A Simple Framework for Masked Image Modeling (1,236 citations)
- Vision Transformer With Deformable Attention (950 citations)
- MixFormer: End-to-End Tracking With Iterative Mixed Attention (843 citations)
- Scaling Vision Transformers (834 citations)
- Mobile-Former: Bridging MobileNet and Transformer (652 citations)
- MAXIM: Multi-Axis MLP for Image Processing (608 citations)
- Scaling Vision Transformers to Gigapixel Images via Hierarchical Self-Supervised Learning (573 citations)
- On the Integration of Self-Attention and Convolution (551 citations)
and 43 more in the interactive map.
RNNs and normalization · 24 papers
- ELIC: Efficient Learned Image Compression With Unevenly Grouped Space-Channel Contextual Adaptive Coding (443 citations)
- GMFlow: Learning Optical Flow via Global Matching (397 citations)
- Practical Stereo Matching via Cascaded Recurrent Network With Adaptive Correlation (339 citations)
- Dataset Distillation by Matching Training Trajectories (236 citations)
- Remember Intentions: Retrospective-Memory-Based Trajectory Prediction (167 citations)
- Towards Efficient and Scalable Sharpness-Aware Minimization (70 citations)
- A Structured Dictionary Perspective on Implicit Neural Representations (68 citations)
- B-Cos Networks: Alignment Is All We Need for Interpretability (65 citations)
- DIP: Deep Inverse Patchmatch for High-Resolution Optical Flow (49 citations)
- MLSLT: Towards Multilingual Sign Language Translation (49 citations)
- DECORE: Deep Compression With Reinforcement Learning (47 citations)
- Deep Equilibrium Optical Flow Estimation (47 citations)
and 12 more in the interactive map.
CNN design · 20 papers
- CMT: Convolutional Neural Networks Meet Vision Transformers (903 citations)
- Image Dehazing Transformer With Transmission-Aware 3D Position Embedding (551 citations)
- CRAFT: Cross-Attentional Flow Transformer for Robust Optical Flow (118 citations)
- Joint Global and Local Hierarchical Priors for Learned Image Compression (112 citations)
- Affine Medical Image Registration With Coarse-To-Fine Vision Transformer (97 citations)
- Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs (81 citations)
- RepMLPNet: Hierarchical Vision MLP With Re-Parameterized Locality (78 citations)
- Learning Optical Flow With Kernel Patch Attention (69 citations)
- A Voxel Graph CNN for Object Classification With Event Cameras (62 citations)
- TVConv: Efficient Translation Variant Convolution for Layout-Aware Visual Processing (41 citations)
- Discrete Time Convolution for Fast Event-Based Stereo (37 citations)
- Efficient Classification of Very Large Images With Tiny Objects (33 citations)
and 8 more in the interactive map.
Neural architecture search · 15 papers
- b-DARTS: Beta-Decay Regularization for Differentiable Architecture Search (112 citations)
- Shapley-NAS: Discovering Operation Contribution for Neural Architecture Search (62 citations)
- Training-Free Transformer Architecture Search (60 citations)
- Global Convergence of MAML and Theory-Inspired Neural Architecture Search for Few-Shot Learning (38 citations)
- SplitNets: Designing Neural Architectures for Efficient Distributed Computing on Head-Mounted Systems (25 citations)
- AutoLoss-Zero: Searching Loss Functions From Scratch for Generic Tasks (22 citations)
- BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule (20 citations)
- ISNAS-DIP: Image-Specific Neural Architecture Search for Deep Image Prior (19 citations)
- Demystifying the Neural Tangent Kernel From a Practical Perspective: Can It Be Trusted for Neural Architecture Search Without Training? (18 citations)
- Neural Architecture Search With Representation Mutual Information (17 citations)
- Performance-Aware Mutual Knowledge Distillation for Improving Neural Architecture Search (16 citations)
- Which Model To Transfer? Finding the Needle in the Growing Haystack (14 citations)
and 3 more in the interactive map.
Pruning and efficient tuning · 12 papers
- Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference (178 citations)
- Revisiting Random Channel Pruning for Neural Network Compression (118 citations)
- CHEX: CHannel EXploration for CNN Model Compression (84 citations)
- Task Adaptive Parameter Sharing for Multi-Task Learning (64 citations)
- Fine-Tuning Image Transformers Using Learnable Memory (53 citations)
- When To Prune? A Policy Towards Early Structural Pruning (49 citations)
- Fire Together Wire Together: A Dynamic Pruning Approach With Self-Supervised Mask Prediction (45 citations)
- Interspace Pruning: Using Adaptive Filter Representations To Improve Training of Sparse CNNs (30 citations)
- DiSparse: Disentangled Sparsification for Multitask Model Compression (23 citations)
- How Well Do Sparse ImageNet Models Transfer? (22 citations)
- Attentive Fine-Grained Structured Sparsity for Image Restoration (15 citations)
- Compressing Models With Few Samples: Mimicking Then Replacing (11 citations)
Quantization · 8 papers
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (116 citations)
- Learnable Lookup Table for Neural Network Quantization (53 citations)
- Instance-Aware Dynamic Neural Network Quantization (39 citations)
- Quantization-Aware Deep Optics for Diffractive Snapshot Hyperspectral Imaging (39 citations)
- Mr.BiQ: Post-Training Non-Uniform Quantization Based on Minimizing the Reconstruction Error (28 citations)
- Data-Free Network Compression via Parametric Non-Uniform Mixed Precision Quantization (17 citations)
- Channel Balancing for Accurate Quantization of Winograd Convolutions (10 citations)
- AlignQ: Alignment Quantization With ADMM-Based Correlation Preservation (4 citations)
Spiking networks · 3 papers
- Event-Based Video Reconstruction via Potential-Assisted Spiking Neural Network (121 citations)
- Training High-Performance Low-Latency Spiking Neural Networks by Differentiation on Spike Representation (120 citations)
- Brain-Inspired Multilayer Perceptron With Spiking Neurons (36 citations)
