From c479f42a4c0a8b4f995a256d42cb95501591b349 Mon Sep 17 00:00:00 2001 From: wassname Date: Sun, 2 Oct 2016 10:19:11 +0800 Subject: [PATCH] update readme --- readme.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/readme.md b/readme.md index ff6b244..da122be 100644 --- a/readme.md +++ b/readme.md @@ -1,11 +1,13 @@ # Pipe segmentation using a convolutional auto-encoder Michael J. Clark + *michael.clark at wassname.org* *Draft 26 September 2016* #### Abstract + Deep convolutional neural networks have been successful in a variety of computer vision tasks, from face detection to handwriting recognition. Convolutional auto-encoders have been particularly successful at object segmentation [[2]](#2)[[3]](#3) in medical imaging, even with small data sets. This paper presents a proof of concept for a new application: pipe detection from aerial drone images. Our model is a convolutional autoencoder with inception blocks, batch-normalisation, strided convolution for downsampling, and skip connections with residual blocks. Data augmentation is utilised to train the model despite the small amount of input data, achieving approximately 73% accuracy. The full implementation (based on keras) and the trained network is available at http://github.com/wassname/pipe-segmentation. ### Introduction