Earthcube’s Recipe for Vehicle Detection

Earthcube’s Recipe for Vehicle Detection
--
Data selection, training, architecture... Dive into the process of designing a deep learning model and get new advices with this handy guide, to develop a powerful vehicle detector on satellite images!
--
CREDIT
Speaker - Julie Imbert
Script - Julie Imbert & Ségolène Husson
With the participation of the Earthcube team
Filmmaker, editing & motion design - Julien Mascheroni
ILLUSTRATIONS
Earthcube Proprietary Detections
Satellite Images
fMoW dataset, Functional Map of the World, CVPR, Gordon Christie, Neil Fendley, James Wilson, and Ryan Mukherjee, 2018
Maxar
Illustrations from Articles
5:44 Olaf Ronneberger, Philipp Fischer, Thomas Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation, 2015, arXiv,
6:19 Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, Deep Residual Learning for Image Recognition, 2015, arXiv,
6:25 Abhijit Guha Roy, Nassir Navab, Christian Wachinger, Concurrent Spatial and Channel Squeeze & Excitation in Fully Convolutional Networks, 2018, arXiv,
8:14 Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry Vetrov, Andrew Gordon Wilson, Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs, 2018, arXiv,
8:15 Andrew Gordon Wilson, Averaging Weights Leads to Wider Optima and Better Generalization, 2018, arXiv.
ARTICLES
Yarin Gal, Riashat Islam, Zoubin Ghahramani, Deep Bayesian Active Learning with Image Data, 2017, arXiv,
Sanghyun Woo, Jongchan Park, Joon-Young Lee, In So Kweon, CBAM: Convolutional Block Attention Module, 2018, arXiv,
Gao Huang, Yixuan Li, Geoff Pleiss, Zhuang Liu, John E. Hopcroft, Kilian Q. Weinberger, Snapshot Ensembles: Train 1, get M for free, 2017, arXiv,
Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry Vetrov, Andrew Gordon Wilson, Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs, 2018, arXiv,
Tsung-Yi Lin, Priya Goyal, Ross B. Girshick, Kaiming He, Piotr Dollár, Focal Loss for Dense Object Detection, 2017, arXiv,
Olaf Ronneberger, Philipp Fischer, Thomas Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation, 2015, arXiv,
Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, Deep Residual Learning for Image Recognition, 2015, arXiv,
Sara Sabour, Nicholas Frosst, Geoffrey E. Hinton, Dynamic Routing Between Capsules, 2017, arXiv,
Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, Jiaya Jia, Pyramid Scene Parsing Network, 2016, arXiv,
Chao Peng, Xiangyu Zhang, Gang Yu, Guiming Luo, Jian Sun, Large Kernel Matters - Improve Semantic Segmentation by Global Convolutional Network, 2017, arXiv,
Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, Identity Mappings in Deep Residual Networks, 2016, arXiv,
Abhijit Guha Roy, Nassir Navab, Christian Wachinger, Concurrent Spatial and Channel Squeeze & Excitation in Fully Convolutional Networks, 2018, arXiv,
Pavel Izmailov, Dmitrii Podoprikhin, Timur Garipov, Dmitry P. Vetrov, Andrew Gordon Wilson, Averaging Weights Leads to Wider Optima and Better Generalization, 2018, arXiv.
Murat Seckin Ayhan, Philipp Berens, Test-time data augmentation for estimation of heteroscedastic aleatoric uncertainty in deep neural networks, 2018,
Yarin Gal, Zoubin Ghahramani, Dropout as a bayesian approximation: Representing model uncertainty in deep learning, 2015.
--
www.earthcube.eu
• AI Academy
/ earthcube

[@da-hn]

I'm loving the series in general, but this video, in particular, was incredibly helpful!

[@RahulPatil-hn9je]

It's best explanation for problem solving in Deep Learning.

[@shreyasdeo3801]

Just Wow!!

[@xavdel0]

Very well explained.
Would you use past data of a place to reduce cost ? For example background who didn't change would not be considered, but now I think about it as you said you got an almost perfect F1-score and doing that would enforce failed classification and accumulate ^^'

[@gren509]

This is basic, non-specific drivel ! Any shcool-kid with an interest will have thought of this. And start with an AI system that makes her "English" understandable !!