图像重建/Low-Level Vision相关的资源整理【持续更新】

图像重建/Low-Level Vision相关的资源整理


整理了图像重建或者Low-Level Vision相关的一些自用资源

学者与机构

  • 清华大学宽带网数字媒体技术实验室的Yebin Liu刘烨斌主页:论文居多是视频相关文章,光场相关的文章主要包括,光场图像处理(包括基本概念理论、采集、超分辨率重构、深度估计、压缩、编码、显示和应用)、Light Field Reconstruction Using Deep Convolutional Network on EPI(提供源码)、Light Field Depth Estimation via Epipolar Plane Image Analysis and Locally Linear Embedding、Light Field Editing Based on Reparameterization、Light Field From Micro-baseline Image Pair等等
  • 东北大学吴高昌:光场图像处理综述(Light Field Image Processing: An Overview)、Light Field Reconstruction Using Convolutional Network on EPI and Extended Applications,(最新一篇)Spatial-Angular Attention Network for Light Field Reconstruction

领域相关的期刊/会议

  • 1、IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE(PAMI)
  • 2、International Journal of Computer Vision(IJCV)
  • 3、IEEE Transactions on Image Processing(TIP)
  • 4、ACM Transactions on Graphics (TOG)
  • 5、IEEE Transactions on Circuits and Systems for Video Technology(TCSVT)
  • 6、IEEE Transactions on Multimedia(TMM)
  • 7、IEEE Conference on Computer Vision and Pattern Recognition(CVPR)

资源

NAS相关

Memory-Efficient Hierarchical Neural Architecture Search for Image Denoising(CVPR2020)

AdversarialNAS: Adversarial Neural Architecture Search for GANs(CVPR2020)

All in One Bad Weather Removal Using Architectural Search(CVPR2020)

AutoGAN: Neural Architecture Search for Generative Adversarial Networks(ICCV2019)

Fast, accurate and lightweight super-resolution with neural architecture search

HNAS: Hierarchical Neural Architecture Search for Single Image Super-Resolution

Efficient Residual Dense Block Search for Image Super-Resolution(AAAI2019)

GNN相关

Cross-Scale Internal Graph Neural Network for Image Super-Resolution(NeurIPS2020)

Transformer相关

Learning Texture Transformer Network for Image Super-Resolution(CVPR2020)

Attention相关

Second-order Attention Network for Single Image Super-resolution (CVPR-2019)

光场相关

Spatial-Angular Attention Network for Light Field Reconstruction(GaochangWu)

Spatial-Angular Interaction for Light Field Image Super-Resolution(ECCV2020)

Light Field Spatial Super-Resolution via Deep Combinatorial Geometry Embedding and Structural Consistency Regularization(CVPR2020)

Learning Fused Pixel and Feature-Based View Reconstructions for Light Fields(CVPR2020)

Attention-based View Selection Networks for Light-field Disparity Estimation(AAAI2020)

全景相关

Panoptic-Based Image Synthesis(CVPR2020)

Latent space/Latent code相关

Bringing Old Photos Back to Life(CVPR2020)

Old Photo Restoration via Deep Latent Space Translation(PAMI Under Review)

Image Processing Using Multi-Code GAN Prior(CVPR2020)

参考

[1] 光场相关
[2] 光场深度估计(Light Field Depth Estimation)

码字不易,如果您觉得有帮助,欢迎点赞和收藏~~

### Event-Based Image Low-Level Processing Techniques Event-based cameras represent a significant shift from traditional frame-based sensors by capturing asynchronous events triggered by changes in light intensity. These devices provide high temporal resolution and low latency data, making them suitable for real-time applications such as robotics and augmented reality. In event-based image processing, several key techniques are employed to handle the unique characteristics of this type of sensor output: #### 1. Event Filtering and Denoising Events generated by these cameras can be noisy due to various factors including environmental lighting conditions or hardware limitations. Advanced filtering algorithms aim at removing noise while preserving important features. For instance, bilateral filters have been adapted specifically for use with event streams[^1]. ```python import numpy as np def bilateral_filter(events, sigma_spatial=0.5, sigma_intensity=0.1): filtered_events = [] for e in events: weight_sum = value_sum = 0 for neighbor in get_neighbors(e): spatial_dist = np.linalg.norm(neighbor.pos - e.pos) intensity_diff = abs(neighbor.intensity - e.intensity) w = np.exp(-spatial_dist**2 / (2*sigma_spatial**2)) * \ np.exp(-intensity_diff**2 / (2*sigma_intensity**2)) weight_sum += w value_sum += w * neighbor.value if weight_sum != 0: filtered_value = value_sum / weight_sum filtered_events.append(Event(pos=e.pos, timestamp=e.timestamp, polarity=int(filtered_value>0))) return filtered_events ``` #### 2. Frame Reconstruction One common task involves reconstructing conventional frames from raw event sequences. This process allows existing vision systems designed around standard imaging modalities to leverage information provided by event-driven sensors without requiring complete redesigns. Methods like backpropagation through time combined with recurrent neural networks facilitate efficient reconstruction tasks[^2]. #### 3. Feature Extraction Extracting meaningful descriptors directly from event clouds enables more robust object recognition under challenging scenarios where traditional methods may fail. Spatiotemporal interest points detected within dense event volumes serve as reliable landmarks that help maintain tracking accuracy even when subjects move rapidly across scenes.
评论 3
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值