General-purpose Convolutional Neural Networks Research Articles

Brain Age Prediction Using Multi-Hop Graph Attention Combined with Convolutional Neural Network.

Convolutional neural networks (CNNs) have been used widely to predict biological brain age based on brain magnetic resonance (MR) images. However, CNNs focus mainly on spatially local features and their aggregates and barely on the connective information between distant regions. To overcome this issue, we propose a novel multi-hop graph attention (MGA) module that exploits both the local and global connections of image features when combined with CNNs. After insertion between convolutional layers, MGA first converts the convolution-derived feature map into graph-structured data by using patch embedding and embedding-distance-based scoring. Multi-hop connections between the graph nodes are modeled by using the Markov chain process. After performing multi-hop graph attention, MGA re-converts the graph into an updated feature map and transfers it to the next convolutional layer. We combined the MGA module with sSE (spatial squeeze and excitation)-ResNet18 for our final prediction model (MGA-sSE-ResNet18) and performed various hyperparameter evaluations to identify the optimal parameter combinations. With 2788 three-dimensional T1-weighted MR images of healthy subjects, we verified the effectiveness of MGA-sSE-ResNet18 with comparisons to four established, general-purpose CNNs and two representative brain age prediction models. The proposed model yielded an optimal performance with a mean absolute error of 2.822 years and Pearson's correlation coefficient (PCC) of 0.968, demonstrating the potential of the MGA module to improve the accuracy of brain age prediction.

Users' photos of items can reveal their tastes in a recommender system

Recommender Systems (RS) are based on the generalization of the observed interactions of a population of users with a collection of items. Collaborative Filters (CF) give good results, but they degrade when there are few interactions to learn from. The alternative would be to observe some features of the users that could be linked to their tastes. However, specific information on users or items is often not available. In this research work, we explore how to exploit the photos of items taken by users. Our aim is to assign similar meanings to the photos of items with which the same group of users interacted. For this purpose, we define a multi-label classification task from images to sets of users. The classifier uses a general-purpose convolutional neural network to extract the basic visual features, followed by additional layers necessary to accomplish the learning task. To evaluate our proposal we compared it with CFs, using two tourism datasets that include: restaurants of six cities and points of interest of three locations. According to the experimentation carried out, the poor results achieved by CFs are outperformed by our proposal, which takes into account the visual and taste semantics of the available photos.

A deep learning approach to automatic gingivitis screening based on classification and localization in RGB photos

Routine dental visit is the most common approach to detect the gingivitis. However, such diagnosis can sometimes be unavailable due to the limited medical resources in certain areas and costly for low-income populations. This study proposes to screen the existence of gingivitis and its irritants, i.e., dental calculus and soft deposits, from oral photos with a novel Multi-Task Learning convolutional neural network (CNN) model. The study can be meaningful for promoting the public dental health, since it sheds light on a cost-effective and ubiquitous solution for the early detection of dental issues. With 625 patients included in this study, the classification Area Under the Curve (AUC) for detecting gingivitis, dental calculus and soft deposits were 87.11%, 80.11%, and 78.57%, respectively; Meanwhile, according to our experiments, the model can also localize the three types of findings on oral photos with moderate accuracy, which enables the model to explain the screen results. By comparing to general-purpose CNNs, we showed our model significantly outperformed on both classification and localization tasks, which indicates the effectiveness of Multi-Task Learning on dental disease detection. In all, the study shows the potential of deep learning for enabling the screening of dental diseases among large populations.

Collaborative General Purpose Convolutional Neural Networks

Collaborative General Purpose Convolutional Neural Networks

Toward making canopy hemispherical photography independent of illumination conditions: A deep-learning-based approach

Hemispherical photography produces the most accurate results when working with well-exposed photographs acquired under diffuse light conditions (diffuse-light images). Obtaining such data can be prohibitively expensive when surveying hundreds of plots is required. A relatively inexpensive alternative is using photographs acquired under direct sunlight (sunlight images). However, this practice leads to high errors since the standard processing algorithms expect diffuse-light imagery. Here, instead of using classification algorithms, which is the unique dominant practice, we approached the processing of sunlight images using deep learning (DL) regression. We implemented DL systems by using the general-purpose convolutional neural networks known as VGGNet 16, VGGNet 19, Res-Net, and SE-ResNet. We trained them with 608 samples acquired in a South American temperate forest populated by Nothofagus pumilio. For their evaluation, we used 113 independent samples. Each sample (X, Y) consisted of one or several sunlight images (X), and the plant area index (PAI) and effective PAI (PAIe) extracted from a diffuse-light image (Y). The sunlight images include clear sky and broken clouds with sun elevation from 15° to 47°. We obtained the best results with the SE-ResNet architecture. The system requires a low-resolution input reprojected to cylindrical, and it can make predictions with 10% root mean square error, even from pictures acquired with automatic exposure, which challenge previous findings. Furthermore, similar results (R2= 0.9, n = 104) can be obtained by feeding the system with photographs acquired with an inexpensive fisheye converter attached to a smartphone. Altogether, results indicate that our approach is a cost-efficient option for surveying hundreds of plots under direct sunlight. Therefore, combining our method with the traditional procedures provides processing solutions for virtually all kinds of illumination conditions.

Avances en foto-identificación automatizada de fauna silvestre

La foto-identificación de especies silvestres es un recurso base para la obtención de información necesaria para diversas tareas de investigaciones biológicas. Hoy en día el crowdsourcing y la ciencia ciudadana están comenzando a desempeñar un rol importante en la recopilación de datos científicos. Esta fuente de datos permite aumentar considerablemente el número de registros en la base de muestreo de diferentes proyectos científicos, especialmente los relacionados con los modelos de capturarecaptura fotográfica de vida silvestre. No obstante, mientras que se aumenta la cantidad de datos recopilados de fuentes no científicas, se presenta un nuevo desafío, el procesamiento masivo de manera ágil y eficiente, que permita limpiar y seleccionar los datos relevantes para las siguientes etapas.Este trabajo aborda la automatización de la primer etapa del proceso de foto-identificación de cetáceos, el cual se trata de la detección de la presencia o ausencia de la región de interés en la imagen (ROI). Para ello, se especializó una red neuronal convolucional de propósito general (Mask R-CNN) con imágenes de delfines de la especie Cephalorhynchus commersonii recolectadas en diferentes sitios de la costa patagónica durante un período de siete años.

Scale insensitive and focus driven mobile screen defect detection in industry

Abstract With the wide-spread of smartphones, mobile phone screen has become an important IO device in HCI and its quality is of great matter in interaction. Traditional defect detection process involves heavy labor cost or relies on unstable low-level features and suffers from both scale and model sensitive problems. Screen defect varies in size, shape, intensity and is hard to be described. Efficient and accurate detection system remains an urgent need in mobile phone screen manufacturing. In this paper, we propose an end-to-end screen defect detection framework. A defect detection network with merging and splitting strategies (MSDDN) to deal with multiple size and shape variations of defect image patches is firstly designed. After training, feature maps of the last layer before the output of MSDDN can be regarded as good representations of a screen image patch. These feature maps are concatenated into a unified feature vector. We then train a recurrent neural network (SCN) to decide which input screen image patch in a sequence is the most likely to contain defects, where the patches are cropped from the same image, and the feature maps are used as input. As SCN emphasizes on the comparison of image patches from one image, it is less sensitive to different screen batches. The patch with the highest probability of containing defects, or called focus area, is further processed with a sliding window to fill in the MSDDN to produce the final results. Finally, to improve the efficiency of the calculation process to fulfill the real industrial demand, we perform both filter selection and weight quantization on the weights in MSDDN under the purpose of building a low-precision version network without great loss in accuracy (MSDDN-l). Experimental results show MSDDN can better handle the defect variations than traditional models and general purpose convolutional neural networks. Meanwhile, SCN can accurately predict the focus area, and MSDDN-l can greatly improve the efficiency.

Caffe con Troll: Shallow Ideas to Speed Up Deep Learning.

We present Caffe con Troll (CcT), a fully compatible end-to-end version of the popular framework Caffe with rebuilt internals. We built CcT to examine the performance characteristics of training and deploying general-purpose convolutional neural networks across different hardware architectures. We find that, by employing standard batching optimizations for CPU training, we achieve a 4.5× throughput improvement over Caffe on popular networks like CaffeNet. Moreover, with these improvements, the end-to-end training time for CNNs is directly proportional to the FLOPS delivered by the CPU, which enables us to efficiently train hybrid CPU-GPU systems for CNNs.