Rethinking the Zigzag Flattening for Image Reading
Sequence ordering of word vector matters a lot to text reading, which has been proven in natural language processing (NLP). However, the rule of different sequence ordering in computer vision (CV) was not well explored, e.g., why the ``zigzag" flattening (ZF) is commonly utilized as a default option to get the image patches ordering in vision networks. Notably, when decomposing multi-scale images, the ZF could not maintain the invariance of feature point positions. To this end, we investigate the Hilbert fractal flattening (HF) as another method for sequence ordering in CV and contrast it against ZF. The HF has proven to be superior to other curves in maintaining spatial locality, when performing multi-scale transformations of dimensional space. And it can be easily plugged into most deep neural networks (DNNs). Extensive experiments demonstrate that it can yield consistent and significant performance boosts for a variety of architectures. Finally, we hope that our studies spark further research about the flattening strategy of image reading.
Code (0)
등록된 구현이 없습니다.
Tasks
Image ClassificationRepresentation LearningSemantic SegmentationMethods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening
In reinforcement learning for large language models, Proximal Policy Optimization (PPO) commonly uses a critic to estimate state values and reduce the variance of policy updates. However, we uncover a systematic failure …
Reinforcement Learning3D zigzag for multislicing, multiband and video processing
We present a 3D zigzag rafter (first in literature) which allows us to obtain the exact sequence of spectral components after application of Discrete Cosine Transform 3D (DCT-2D) over a cube. Such cube represents part of…
Consistent Story Generation with Asymmetry Zigzag Sampling
Text-to-image generation models have made significant progress in producing high-quality images from textual descriptions, yet they continue to struggle with maintaining subject consistency across multiple images, a fund…
Image GenerationStory GenerationStory VisualizationText to Image Generation+2Z-GCNETs: Time Zigzags at Graph Convolutional Networks for Time Series Forecasting
There recently has been a surge of interest in developing a new class of deep learning (DL) architectures that integrate an explicit time dimension as a fundamental building block of learning and representation mechanism…
Time SeriesTime Series AnalysisTime Series ForecastingComplexity of zigzag sampling algorithm for strongly log-concave distributions
We study the computational complexity of zigzag sampling algorithm for strongly log-concave distributions. The zigzag process has the advantage of not requiring time discretization for implementation, and that each propo…