paper: DilatedNet: MULTI-SCALE CONTEXT AGGREGATION BY DILATED CONVOLUTIONS | [ code/pytorch_code ]

paper: ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation

paper: U-Net: Convolutional Networks for Biomedical Image Segmentation

paper: Going deeper with convolutions

Inception-V1 GoogLeNet

paper: Fully Convolutional Networks for Semantic Segmentation

paper: Systematic evaluation of CNN advances on the ImageNet

paper: SQUEEZENET: ALEXNET-LEVEL ACCURACY WITH 50X FEWER PARAMETERS AND <0.5MB MODEL SIZE with pytorch code

1.install Ubuntu18.04 and update source

Update to mirrors.aliyun.com, reference this in step 1.

paper: Fast R-CNN

SPPnet solved the R-CNN’s problem that it extracts features for each of the 2k~ region proposal and costs a lot of time. SPPnet runs the convolutional layers only once on the entire image (regardless of the number of windows), and then extract features by SPP-net on the feature maps. But it is not an end-to-end model so that extracted features need to be written to the disk, and uses two stages for classification and bbox regression. Fast R-CNN can do back-propagation end-to-end.

paper: VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION

paper: Selective Kernel Networks

It is well-known in the neuroscience community that the receptive field size of visual cortical neurons are modulated by the stimulus, which has been rarely considered in constructing CNNs. In Selective Kernel (SK) unit, multiple branches with different kernel sizes are fused using softmax attention that is guided by the information in these branches.

paper: Network In Network

Contributions

paper: ImageNet Classification with Deep Convolutional Neural Networks