上海科技大学信息科学与技术研讨会（SSIST day3）笔记

一、Deep Network and Beyond

presenters: Alan L. Yuille Johns Hopkins University
1.More Structured Attacks:Occlusion
Jianyu Wang et al. Annals of Mathematical Sciences. 2018
2.DeepVote
DeepVote is an end-to-end Deep Net which is robusts to occluders.
Zhishuai Zhang, Cihang Xie,Jianyu Wang et al. CVPR. 2018
3.Stress Testing Vision Algorithms: UnrealCV
UnrealCV : Weichao Qiu https://unrealcv.org/

二、Deep Visual Patterns Beyond Recognition

Presenters: Gang Hua Microsoft
1.Visualizing and Understanding CNN
Zeiler and Ferqus, Visualizing and Understanding Convolutional Networks (ECCV 2014)
2.Neural style transfer
Gatys et al, A Neural Algorithm of Artistic Style (CVPR 2016)
3.StyleBank: an explicit representation of style
Chen et al, StyleBank:An explicit Representation for Neural Image Style Transfer (CVPR 2017)
4.Spatial coherent video stylization
Chen et al, Coherent Online Video Style Transfer (ICCV 2017)
5.Deep image analogy
Liao et al, Visual Attribute Transfer through Deep Image Analogy (SIGGRAPH 2017)
6.Generative adversarial networks
Goodfellow et al Generative Adversarial Nets (NIPS 2014)
7.CVAE-GAN
Bao et al, CVAE-GAN: Fine-Grained Image Generation through Asymetric Training (ICCV 2014)
8.Openset
Bao et al, Towards Open-set Identity Preserving Face Synthesis (CVPR 2018)

三、Video Content Analysis With Deep Learning

presenters: Xiangyang Xue Fudan University
1.Computer Visual



2.Fudan-Columbia Video Dataset (FCVID)
Exploiting Feature and Class Relationships in Video Categorization with Regularized Deep Neural Networks (IEEE TPAMI 2018)
3.A representation work: Two-Stream CNN
K.Simonyan and A.Zisserman , Two-stream convolutional networks for action recognition in videos (NIPS 2014)
4.A Hybrid Deep Learning Framework
Modeling Spatial-Temporal Clues in a Hybrid Deep Learning Framework for Video Classification (ACM Multimedia 2015)
5. Adaptive Multimodal Fusion
Multi-Stream Multi-Class Fusion of Deep Networks for Video Classification (ACM Multimodal 2016)
6.Weakly Supervised Dense Video Captioning
Weakly Supervised Dense Video Captioning (CVPR 2017)
7.Learning deeply supervised object detectors from scratch
Dsod: Learning deeply supervised object detectors from scratch (ICCV 2017)
8.Person Re-id with salient multi-scale features
Multi-scale Deep Learning Architectures for Person Re-identification (ICCV 2017)
9.Dual Skipping Networks
Dual Skipping Networks (CVPR 2018)

四、Deep Learning :Core Networks,Industry Applications, and Future Directions

presenters: Shuicheng Yan Qihoo 360
1.Network in Network : bring 1*1 convolution to the community
Densely Connected Convolutional Networks (CVPR 2017)
2.Dual Path Architecture
Aggregated residual transformations for deep neural networks (CVPR 2017)
3. ILSVRC 2017 : Object Localization & Detection
R-CNN: Towards Faster real-time object detection with region proposal networks (NIPS 2015)
4.Multi-fiber Networks (w/ Transistor)

五、Towards Human-level Visual Computing: Present and Challenges

Presenters: Jian Sun Face++
1. Face++ = Megvii
https://www.faceplusplus.com/
2.Representation - Part
Xudong Cao, Yichen Wei, Fang Wen, and Jian Sun. Face Alignment by Explicit Shape Regression (CVPR 2012)
P.Felzenszwalb, R.Girshick, D.McAllester , D.Ramanan. Object Detection with Discriminatively Trained Part Base Models. (PAMI 2010)
3.Representation - Early Learning-based
Viola & Jones. Robust Real-time Object Detection (IJCV 2001)
4.Representation - Feature-based eg:SIFT/HOG
David Low.Distinctive Image Features from Scale-Invariant Keypoints (IJCV 2004）
5.Representation - Feature-based eg:for Face
Blessing of Dimensionality:High-dimensional Feature and Its Efficient Compression for Face Verification (CVPR 2013)
6.Deep Convolutional Neural Networks
ImageNet Classification with Deep Convolutional Neural Networks (NIPS 2012)

对Resnet的不同理解
1.Un-folded RNN view
Bridging the Gaps Between Residual Learning,Recurrent Neural Networks and Visual Cortex. (ArXiv 2016)
2.”Ensembles” View
Residual Networks Behave Like Ensembles of Relatively Shallow Networks (NIPS 2017)
3. Iterative Refinement View
Residual Connections Encourage Iterative Inference (ArXiv 2018)
4.Loss Surface View
Visualizing The Loss Landscape Of Neural Nets (ICLR 2018)

各大网络的发展

1.ShuffleNet v2
ShuffleNet v2:Light-weight CNN Architecture Design for Fast Inference (arXiv 2018)
2.Low-bit Networks
Quantizing deep convolutional networks for efficient inference: A whitepaper (arXiv:1806.08342 2018 )
3.DorefaNet
DoReFa-Net:Training Low Bitwidth Convolutional Neural with Low Bitwidth Gradients (arXiv 1606.06160 2016)
4.Object Detection:Rapid Progresses
http://host.robots.ox.ac.uk/pascal/VOC/
5.MegDet
A Large Mini-Batch Object Detector (CVPR 2018 ,arXiv:1711.07240)
6.CrowdHuman
A Benchmark for Detecting Human in a Crowd
链接：http://arxiv.org/pdf/1805.00123.pdf