plateau lr policy 好久没有冒泡了,哈哈。plateau lr policy来自于PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection这篇论文。这里从issue里面整理出来的,欢迎大家指点,以及如果你用了plateau lr policy来做实验,是否可以告知下效果如何?issues:Add "plate
SSD系列 1SSD: Single Shot MultiBox Detector. ECCV16 (arxiv)2R-SSD: enhancement of ssd by concatenating feature maps for object detection. BMVC17. (arxiv)3DSSD: Deconvolut
human pose regression by combining indirect part detection and contextual information - arxiv - 1710 好久没写了,心血来潮好吧,怎么简单怎么来:human pose regression by combining indirect part detection and contextual information - arxiv - 1710.02322论文链接该论文的亮点在于重新用回了regression,而不是直接基于heat map的part detection方法。
Improving Facial Attribute Prediction using Semantic Segmentation, in arxiv 2017.04 笔者最近在关注fine-grained方面的paper,发现有以下的方面去做:1 part-based2 weakly-supervised的,如second-orderless pooling(Compact Bilinear Pooling)等3 还是weakly-supervised的,但用上了proposals/grids/regions(如selective search)等
Residual Attention Network for Image Classification, cvpr17 人至懒则无敌。cvpr17的论文,很有意思,值得一读和复现。(笔者懒,还是坐等开源吧)Residual Attention Network for Image Classification. In CVPR 2017.该模型的结构参考了: Stacked hourglass networks for human pose estimation. In arXiv 2016. (笔者
Convolutional Low-Resolution Fine-Grained Classification. In arXiv, 2017.03 Convolutional Low-Resolution Fine-Grained Classification. In arXiv, 2017.03论文链接:Convolutional Low-Resolution Fine-Grained Classification.论文的idea很简单,就是把super resolution和classifier嵌入到一个模型中,并en
Pyramid Scene Parsing Network - arxiv - 1612.01105 Pyramid Scene Parsing Network - arxiv - 1612.011052016年imagenet的scene parsing的冠军github: https://github.com/hszhao/PSPNetproject: http://appsrv.cse.cuhk.edu.hk/~hszhao/projects/pspnet/index.h
Fully Convolutional Instance-aware Semantic Segmentation - arXiv 16.11 Fully Convolutional Instance-aware Semantic Segmentation - arXiv 16.11论文链接:https://arxiv.org/pdf/1611.07709v1.pdfgithub:https://github.com/daijifeng001/TA-FCN=====参考博文:1Instance-sensitive Fu
Aggregated Residual Transformations for Deep Neural Networks - arxiv 16.11 Aggregated Residual Transformations for Deep Neural Networks, arXiv 16.11.论文链接: https://arxiv.org/abs/1611.05431=====这个是全明星阵容的paper,提出了深度模型另一个重要的需要考量的纬度:cardinality。(用简单的方式做不简单的事情)众所周知,提高模
speed improvement by merging batch normalization and scale 最近在看“在测试的时候, 怎么merge BN 和 Scale layer到conv layer”来加速。具体可以看:speed improvement by merging batch normalization and scale #5
Deep Image Retrieval: Learning global representations for image search. In ECCV, 2016. Deep Image Retrieval: Learning global representations for image search. In ECCV, 2016.论文地址:https://arxiv.org/abs/1604.01325extended version:end to end learning of deep visual representations for i
A Discriminative Feature Learning Approach for Deep Face Recognition, ECCV16. A Discriminative Feature Learning Approach for Deep Face Recognition, ECCV16.论文:http://ydwen.github.io/papers/WenECCV16.pdf代码(caffe):https://github.com/kpzhang93/caffe-face代码(mxnet):https://gith
StuffNet: Using ‘Stuff’ to Improve Object Detection - arxiv 16.10 StuffNet: Using ‘Stuff’ to Improve Object Detection, Arxiv 16.10论文地址:https://arxiv.org/pdf/1610.05861v1.pdf类似的论文:Contextual Priming and Feedback forFaster R-CNN, eccv16.=====好像现在都喜欢用multi-task
PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection - arxiv 2016.08 PVANET: Deep but Lightweight Neural Networks forReal-time Object Detection, arxiv 16.08论文地址:https://arxiv.org/pdf/1608.08021v1.pdf先看leaderboard再看下speed厉害的不要不要的。=====转正题,下面介绍下这篇
Instance-sensitive Fully Convolutional Networks - eccv 2016 Instance-sensitive Fully Convolutional Networks - eccv 2016=====论文地址:http://arxiv.org/abs/1603.08678=====一句话概括: 根据local coherence的特性,以sliding window的方式,利用FCN产生positive-sensitive的instance-le
R-FCN: Object Detection via Region-based Fully Convolutional Networks (NIPS 16), Arxiv 16.05 R-FCN: Object Detection via Region-based Fully Convolutional Networks, Arxiv 16.05=====blogs:论文阅读]R-FCN: Object Detection via Region-based Fully Convolutional Networks: http://blog
- UnitBox An Advanced Object Detection Network,arxiv 16.08 - UnitBox An Advanced Object Detection Network,arxiv 16.08 (download) 该论文提出了一种新的loss function:IoU loss。这点比较有意思,也容易复现。 ====== 论文分析了faster-rcnn和densebox的优缺点: 1 fast
Bootstrapping Face Detection with Hard Negative Examples,arxiv 16.08 - Bootstrapping Face Detection with Hard Negative Examples,arxiv 16.08(download) 昨天在微博上看到这个论文的时候,很是惊讶,为啥?因为出自小米的为发烧而生(跑分) ===== 论文比较短,主要提出了一个hard negative mining的方式,以及基于faste
Multi-view Face Detection Using Deep Convolutional Neural Networks, ICMR 15. Multi-view Face Detection Using Deep Convolutional Neural Networks, ICMR 15. 将Alexnet转成FCN的形式也就是227*227的输入会输出1*1的heat map),将detection问题转为classification问题。 很简单,对每个region/propos