Count-ception: Counting by Fully Convolutional Redundant Counting

Abstract

Counting objects in digital images is a process that should be replaced by machines. This tedious task is time consuming and prone to errors due to fatigue of human annotators. The goal is to have a system that takes as input an image and returns a count of the objects inside and justification for the prediction in the form of object localization. We repose a problem, originally posed by Lempitsky and Zisserman, to instead predict a count map which contains redundant counts based on the receptive field of a smaller regression network. The regression network predicts a count of the objects that exist inside this frame. By processing the image in a fully convolutional way each pixel is going to be accounted for some number of times, the number of windows which include it, which is the size of each window, (i.e., 32x32 = 1024). To recover the true count we take the average over the redundant predictions. Our contribution is redundant counting instead of predicting a density map in order to average over errors. We also propose a novel deep neural network architecture adapted from the Inception family of networks called the Count-ception network. Together our approach results in a 20% relative improvement (2.9 to 2.3 MAE) over the state of the art method by Xie, Noble, and Zisserman in 2016. Wepropose a novel construction of networks and train-arXiv:1703.08710v2 [cs.CV] 23 Jul 2017 1. Pre-process image by padding 2. Process image in a fully convolutional way 3. Combine all counts together into total count for image

UDCT: Unsupervised data to content transformation with histogram-matching cycle-consistent generative adversarial networks

“…The pixels are then summed and divided by the average synthetic cell area to produce a count. This is similar to Cohen et al 's [19] state-of-the-art approach, with the distinction that ours is unsupervised. In a second dataset depicted in Fig.2b) we simply create round-shaped cells of varying intensity in the blue channel, a two-dimensional gaussian map with a standard deviation of 3 in the green and 1 in the red channel.…”

“…The red and green channels encode the local angle ϕ of the worm using [cos(2 · ϕ)/2 + 1, sin(2 · ϕ)/2 + 1]. The results of a modified dataset, where the total number of Table 2: Prediction error for neurons: We compare the mean relative error of both the count-ception network as proposed by Cohen et al [19] and our cycleGAN. The first two rows show the average count and the corresponding standard deviation of both dead and live neurons for each of the three experts.…”

“…Details of the post-processing steps are given in the Supplementary Information. Table 1: Summary on the VGG dataset: UDCT performs slightly worse than Count-ception [19] regardless of which synthetic dataset we use (See Fig.2 a) or b)). However our method is unsupervised and does not directly minimize the mean absolute error during training.…”

Machine Learning based histology phenotyping to investigate epidemiologic and genetic basis of adipocyte morphology and cardiometabolic traits

“…Being able to interrogate high-dimensional endophenotypes in a GWAS framework may yield a more rapid uncovering of genetic variants directly linked to the biological mechanisms that underpin clinically-measured outcomes. Many such methods to derive phenotypes from images are currently being developed 19,22,37,38 .…”

CellCountCV – a web-application for accurate cell counting and automated batch processing of microscopy images using fully-convolutional neural networks

“…Here we describe the development of a deep neural network for quantitative analysis of microscopic images of cells, expressing the ER stress biosensor XBP1-TagRFP. Cell counting was realized with an approach based on the paper by Cohen et al (7) aimed to a very similar research field. The authors proposed a way to count multiple small objects (as compared to the image size) pertaining to a same category using fully-convolutional neural networks (FCNN), which predict the so-called redundant count maps.…”