paper_summary
训练 Trick
[1] learnning rate scheduler: Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
[2] Adam + L2 regularization 会耦合效果差于sgd: DECOUPLED WEIGHT DECAY REGULARIZATION;知乎文章：都9102年了，别再用Adam + L2 regularization了
[3] Adam to SGD 训练过程中Adam转换城SGD: Improving Generalization Performance by Switching from Adam to SGD
Network
[1] Unet: U-Net: Convolutional Networks for Biomedical Image Segmentation
[2] Linknet based Unet: LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation
[3] D-Linknet based Unet: D-LinkNet: LinkNet with Pretrained Encoder and Dilated Convolution for High Resolution Satellite Imagery Road Extraction
Loss
[1] Focal Loss: Focal Loss for Dense Object Detection
[2] 深入比较了主流loss Deep Semantic Segmentation of Natural and Medical Images:A Review
[3] Combo loss Combo Loss: Handling Input and Output Imbalance in Multi-Organ Segmentation
ceter line extraction
[1] DeepCenterline: a Multi-task Fully Convolutional Network for Centerline Extraction
[2] Coronary Artery Centerline Extraction in Cardiac CT Angiography Using a CNN-Based Orientation Classifier
Idea
[1] Geoffrey Hinton Idea(只提出了idea以及可能的新的研究方向没有具体做法): How to represent part-whole hierarchies
待分分类
[1] MEMORY NETWORKS
[2] opening the black box of deep neural network via information
