CV Code | 计算机视觉开源周报 20190604期

点击我爱计算机视觉标星，更快获取CVML新技术

---

本周有不少好的开源代码，含有多个热门主题，含目标检测、实例分割、视觉问答、网络压缩与减枝、医学图像处理分割、去马赛克、目标跟踪、对抗学习。

其中不少已获大量关注，比如谷歌的基于学习的数据增广策略这几天就很火，Cascade R-CNN的升级版也大放异彩。

值得一提的是，医学影像的处理识别领域有四篇文章开源，这是视觉的一大特殊应用领域，也是技术落地造福人类的体现。

目标检测

学习用于目标检测的数据增广策略，COCO数据集50.7 AP

谷歌

Learning Data Augmentation Strategies for Object Detection

Barret Zoph, Ekin D. Cubuk, Golnaz Ghiasi, Tsung-Yi Lin, Jonathon Shlens, Quoc V. Le

https://arxiv.org/abs/1906.11172v1

https://github.com/tensorflow/tpu/tree/master/models/official/detection

Cascade R-CNN再扩展，目标检测COCO数据集50.9 AP，实例分割超越Mask RCNN

Cascade R-CNN: High Quality Object Detection and Instance Segmentation

Zhaowei Cai, Nuno Vasconcelos

https://arxiv.org/abs/1906.09756v1

https://github.com/zhaoweicai/cascade-rcnn

https://github.com/zhaoweicai/Detectron-Cascade-RCNN

介绍：Cascade R-CNN升级！目标检测制霸COCO，实例分割超越Mask R-CNN

实例分割

发明一种聚类损失函数，改进无候选目标的实例分割，速度更快，达到10fps，精度超越Mask RCNN 5%，SOTA

Instance Segmentation by Jointly Optimizing Spatial Embeddings and Clustering Bandwidth

Davy Neven, Bert De Brabandere, Marc Proesmans, Luc Van Gool

https://arxiv.org/abs/1906.11109v1

https://github.com/davyneven/SpatialEmbeddings

视觉问答

用于视觉问答的深度模块协注意力网络

Deep Modular Co-Attention Networks for Visual Question Answering

Zhou Yu, Jun Yu, Yuhao Cui, Dacheng Tao, Qi Tian

https://arxiv.org/abs/1906.10770v1

https://github.com/MILVLG/mcan-vqa

减少视觉问答问题中的单峰偏差

RUBi: Reducing Unimodal Biases in Visual Question Answering

Remi Cadene, Corentin Dancette, Hedi Ben-younes, Matthieu Cord, Devi Parikh

https://arxiv.org/abs/1906.10169v1

http://github.com/cdancette/rubi.bootstrap.pytorch

网络压缩与减枝

网络减枝中的重要性估计，在ResNet-101网络中, 在ImageNet数据集上，取得了 40% FLOPS 降低去除了 30% 参数量, 而只损失了 0.02% 的 top-1 精度 

Importance Estimation for Neural Network Pruning

Pavlo Molchanov, Arun Mallya, Stephen Tyree, Iuri Frosio, Jan Kautz

https://arxiv.org/abs/1906.10771v1

https://github.com/NVlabs/Taylor_pruning

定制深度模型压缩，使用基于正则化相关的滤波级减枝

COP: Customized Deep Model Compression via Regularized Correlation-Based Filter-Level Pruning

Wenxiao Wang, Cong Fu, Jishun Guo, Deng Cai, Xiaofei He

https://arxiv.org/abs/1906.10337v1

https://github.com/ZJULearning/COP

医学图像处理与识别

医学图像分割 | 联合CNN分割与形态学操作的全自动肝衰竭区域估计

Fully Automatic Liver Attenuation Estimation Combing CNN Segmentation and Morphological Operations

Yuankai Huo, James G. Terry, Jiachen Wang, Sangeeta Nair, Thomas A. Lasko, Barry I. Freedman, J. Jeffery Carr, Bennett A. Landman

https://arxiv.org/abs/1906.09549v1

https://github.com/MASILab/ALARM

基于卷积神经网络的显微图像急性淋巴细胞白血病分类

Acute Lymphoblastic Leukemia Classification from Microscopic Images using Convolutional Neural Networks

Jonas Prellberg, Oliver Kramer

https://arxiv.org/abs/1906.09020v1

https://github.com/jprellberg/isbi2019cancer

边界和熵驱动的对抗学习，用于眼底图像分割

Boundary and Entropy-driven Adversarial Learning for Fundus Image Segmentation

Shujun Wang, Lequan Yu, Kang Li, Xin Yang, Chi-Wing Fu, Pheng-Ann Heng

https://arxiv.org/abs/1906.11143v1

https://github.com/EmmaW8/BEAL

精细分割 R-CNN：早产儿点状白质病变分割的两阶段卷积神经网络

Refined-Segmentation R-CNN: A Two-stage Convolutional Neural Network for Punctate White Matter Lesion Segmentation in Preterm Infants

Yalong Liu, Jie Li, Ying Wang, Miaomiao Wang, Fan Wu, Zhicheng Jiao, Jian Yang, Xingbo Gao

https://arxiv.org/abs/1906.09684v1

https://github.com/YalongLiu/Refined-Segmentation-R-CNN

目标跟踪

用于目标跟踪的可微闭合解的特征学习，在OTB2015, NFS, TrackingNet, GOT10k, VOT2018, 和 VOT2019 数据集上达到了SOTA，同时速度30FPS

Learning Features with Differentiable Closed-Form Solver for Tracking

Linyu Zheng, Ming Tang, JinqiaoWang, Hanqing Lu

https://arxiv.org/abs/1906.10414v1

（代码将开源，还未公布地址）

去马赛克

一种轻量级CNN用于去马赛克，速度很快而精度达到最高

Cross-Channel Correlation Preserved Three-Stream Lightweight CNNs for Demosaicking

Niu Yan, Jihong Ouyang

https://arxiv.org/abs/1906.09884v1

（该文出版后，代码将开源）

GAN

使用生成对抗网络（GAN）来研究认知属性，如记忆力，美学和情感效价

MIT、鲁汶大学

GANalyze: Toward Visual Definitions of Cognitive Image Properties

Authors: Lore Goetschalckx (1 and 2), Alex Andonian (1), Aude Oliva (1), Phillip Isola (1) ((1) MIT, (2) KU Leuven)

https://arxiv.org/abs/1906.10112v1

https://github.com/LoreGoetschalckx/GANalyze

神经架构搜索&对抗攻击

将抵御对抗攻击作为神经架构搜索的目标，不断寻找对攻击更加鲁棒的网络结构

Evolving Robust Neural Architectures to Defend from Adversarial Attacks

Danilo Vasconcellos Vargas, Shashank Kotyan

https://arxiv.org/abs/1906.11667v1

http://bit.ly/RobustArchitectureSearch

半监督学习

自监督网络的半监督学习，在SVHN, CIFAR-10 和 CIFAR-100数据集上相较之前的SOTA取得可比较或者更好的结果

Semi-Supervised Learning with Self-Supervised Networks

Phi Vu Tran

https://arxiv.org/abs/1906.10343v1

https://github.com/vuptran/sesemi

---

加群交流

关注计算机视觉与机器学习技术，欢迎加入52CV群，扫码添加CV君拉你入群，

（请务必注明:52CV）

喜欢在QQ交流的童鞋，可以加52CV官方QQ群：702781905。

（不会时时在线，如果没能及时通过验证还请见谅）

长按关注我爱计算机视觉