深度学习的应用(Applications of Deep Learning).pptx

Applications of Deep Learning Content: 1. Introduction of Deep Learning 2. Deep Learning for classification 3. Deep Learning for Low Level Vision Image Restoration Image Enhancement 4. Deep Learning for High Level Vision Visual Tracking Semantic Segmentation 5. Other Applications How to solve? Back propagation vanishing/exploding gradients! Root reason: too many weights? Solutions for vanishing/exploding gradients! Vanishing gradients Pre-train Batch input Dropout Active function CNN Structure of Network What is Deep Learning? More deeper More powerful representation ability! 89 ImageNet LSVRC contest 2012 NIPS Alex et al. ImageNet Classification with Deep Convolutional Neural Networks 2015 ICLR Karen et al. Very Deep Convolutional Networks for Large-Scale Image Recognition VGG 19-Layers 2015 CVPR Christian et al. Going Deeper with Convolutions GoolLeNet 22-Layers ImageNet LSVRC contest 2015 arXiv He et al. Deep Residual Learning for Image Recognition 152-Layers ImageNet LSVRC contest 2015 arXiv He et al. Deep Residual Learning for Image Recognition 152-Layers Image Restoration Image denoise 2009 NIPS Viren et al. Natural Image Denoising with Convolutional Networks 5-Layers 2012 CVPR Harold et al. Image denoising: Can plain Neural Networks compete with BM3D? 3-Layers Image Restoration Image deblur 2014 NIPS Xu et al. Deep Convolutional Neural Network for Image Deconvolution Image Restoration Image super-resolution 2015 TPAMI Dong et al. Image Super-Resolution Using Deep Convolutional Networks Image Restoration Image inpainting 2015 NIPS Xu et al. Shepard Convolutional Neural Networks Image Restoration Image inpainting Significance: Shadow removal Highlight etc. Not mentioned: Salience Dirt/rain removal Image Restoration Image enhancement 2015 ICCV Cheng et al. Deep Colorization Train set: extract features, not learn features Low-level patch feature Mid-level DAISY feature Global features (gist) High level vision Visual tracking 2015 ICCV Wang et al. Visual Tracking