From: Foivos Diakogiannis <[email protected]> Date: Saturday, November 14, 2020 at 9:32 PM To: apachemxnetday <[email protected]> Cc: Foivos Diakogiannis <[email protected]> Subject: Application for presentation
Dear all, I would please like to be considered for presenting on the apache mxnet day, in the category "Research and applications". I assume that the mxnet day will also include applications written with apache mxnet that deviate from computer science - hence my application. If yes, then you may find the topic of my presentation interesting: I will present our latest work on change detection ( https://arxiv.org/abs/2009.02062 - under review, https://github.com/feevos/ceecnet ). The talk will be focused on change detection, with some highlights from our previous research on semantic segmentation (resunet-a https://www.sciencedirect.com/science/article/pii/S0924271620300149 , https://github.com/feevos/resuneta), all built with Apache mxnet. Title: Looking for change? Roll the Dice and demand Attention Abstract: Change detection, i.e. identification per pixel of changes for some classes of interest from a set of bi-temporal co-registered images,is a fundamental task in the field of remote sensing. It remains challenging due to unrelated forms of change that appear at different times in input images. These are changes due to different environmental conditions or simply changes of objects that are not of interest. Here, we propose a reliable deep learning framework for the task of semantic change detection in very high-resolution aerial images. Our framework consists of a new loss function, new attention modules, new feature extraction building blocks, anda new backbone architecture that is tailored for the task of semantic change detection. Specifically, we define a new form of set similarity, that is based on an iterative evaluation of a variant of the Dice coefficient. We use this similarity metric to define a new loss function as well as a new spatial and channel convolution Attention layer (the FracTAL). The new attention layer, designed specifically for vision tasks, is memory efficient, thus suitable for use in all levels of deep convolutional networks. Based on these,we introduce two new efficient self-contained feature extraction convolution units. We term these units CEECNet and FracTAL ResNet units. We validate the performance of these feature extraction building blocks on the CIFAR10 reference data and compare the results with standard ResNet modules. Further, we introduce a new encoder/decoder scheme, a network macro-topology, that is tailored for the task of change detection. We validate our approach by showing excellent performance and achieving state of the art score (F1 and Intersection over Union - hereafter IoU) on two building change detection datasets, namely, the LEVIRCD (F1:0.918, IoU: 0.848) and the WHU (F1: 0.938, IoU: 0.882) datasets Kind regards, Foivos I. Diakogiannis ---------------------------------------------------------------------------- Dr. Foivos I. Diakogiannis Senior Research Fellow - Data Science ICRAR - The University of Western Australia P: +61 8 6488 7206 E: [email protected]<mailto:[email protected]>
