Dietary intake, the process of determining what someone eats during the course of a day, provides valuable insights for mounting intervention programs for prevention of many chronic diseases such as obesity and cancer. The goals of the Technology Assisted Dietary Assessment (TADA) System, developed at Purdue University, is to automatically identify and quantify foods and beverages consumed by utilizing food images acquired with a mobile device. Color correction serves as a critical step to ensure accurate food identification and volume estimation. We make use of a specifically designed color checkerboard (i.e. a fiducial marker) to calibrate the imaging system so that the variations of food appearance under different lighting conditions can be determined. In this paper, we propose an image quality enhancement technique by combining image de-blurring and color correction. The contribution consists of introducing an automatic camera shake removal method using a saliency map and improving the polynomial color correction model using the LMS color space.
Abstract The use of image‐based dietary assessment methods shows promise for improving dietary self‐report among children. The Technology Assisted Dietary Assessment (TADA) food record application is a self‐administered food record specifically designed to address the burden and human error associated with conventional methods of dietary assessment. Users would take images of foods and beverages at all eating occasions using a mobile telephone or mobile device with an integrated camera [e.g. Apple iP hone, Apple iP od Touch (Apple Inc., Cupertino, CA, USA); Nexus One (Google, Mountain View, CA, USA)]. Once the images are taken, the images are transferred to a back‐end server for automated analysis. The first step in this process is image analysis (i.e. segmentation, feature extraction and classification), which allows for automated food identification. Portion size estimation is also automated via segmentation and geometric shape template modeling. The results of the automated food identification and volume estimation can be indexed with the Food and Nutrient Database for Dietary Studies to provide a detailed diet analysis for use in epidemiological or intervention studies. Data collected during controlled feeding studies in a camp‐like setting have allowed for formative evaluation and validation of the TADA food record application. This review summarises the system design and the evidence‐based development of image‐based methods for dietary assessment among children.
Food image classification is a fundamental step of image-based dietary assessment, enabling automated nutrient analysis from food images. Many current methods employ deep neural networks to train on generic food image datasets that do not reflect the dynamism of real-life food consumption patterns, in which food images appear sequentially over time, reflecting the progression of what an individual consumes. Personalized food classification aims to address this problem by training a deep neural network using food images that reflect the consumption pattern of each individual. However, this problem is under-explored and there is a lack of benchmark datasets with individualized food consumption patterns due to the difficulty in data collection. In this work, we first introduce two benchmark personalized datasets including the Food101-Personal, which is created based on surveys of daily dietary patterns from participants in the real world, and the VFNPersonal, which is developed based on a dietary study. In addition, we propose a new framework for personalized food image classification by leveraging self-supervised learning and temporal image feature information. Our method is evaluated on both benchmark datasets and shows improved performance compared to existing works. The dataset has been made available at: https://skynet.ecn.purdue.edu/~pan161/dataset_personal.html
Continual learning in online scenario aims to learn a sequence of new tasks from data stream using each data only once for training, which is more realistic than in offline mode assuming data from new task are all available. However, this problem is still under-explored for the challenging class-incremental setting in which the model classifies all classes seen so far during inference. Particularly, performance struggles with increased number of tasks or additional classes to learn for each task. In addition, most existing methods require storing original data as exemplars for knowledge replay, which may not be feasible for certain applications with limited memory budget or privacy concerns. In this work, we introduce an effective and memory-efficient method for online continual learning under class-incremental setting through candidates selection from each learned task together with prior incorporation using stored feature embeddings instead of original data as exemplars. Our proposed method implemented for image classification task achieves the best results under different benchmark datasets for online continual learning including CIFAR-10, CIFAR-100 and CORE-50 while requiring much less memory resource compared with existing works.
Texture synthesis is the process of generating a large texture image from a small texture sample. The synthesized image must appear as though it has the same underlying structural content as the input texture sample. However, most texture synthesis methods require the user to tune parameters for different input or provide feedback to the system to achieve satisfactory results. To make texture synthesis approaches more efficient and user friendly, we propose a fully automatic method to select a set of suitable parameters for texture synthesis that can be applied on commonly used textures. Our method uses Convolutional Neural Network (CNN) to predict the optimal parameters for texture synthesis based on image quilting algorithm . Our method showed satisfactory results on different types of textures.
In this paper we present a video coding approach similar to texture- based methods but based on motion models. We consider motion perception properties instead of spatial texture properties of the video sequence. We integrate a motion classification algorithm to separate foreground objects containing noticeable motion from the background. These background areas are labeled as skipped areas that are not encoded. After decoding, frame reconstruction is performed by inserting the skipped background into the decoded frames. We are able to show as much as 15% an improvement over previous texture- based implementations in terms of video compression efficiency.
American adolescents have a nutrient-poor diet pattern, which is particularly high in added sugars, putting them at risk for obesity and type 2 diabetes (T2D). We aimed to assess dietary intake of added sugars in adolescents and relationships with glycemia and body mass index (BMI). Cross-sectional, baseline measures were obtained from an ongoing, randomized controlled behavioral intervention to prevent adolescent T2D. Participants, using the Technology Assisted Dietary Assessment system (TADA), created a mobile, imaged-based, four-day food record which the Nutrition Data System for Research (NDSR, University of Minnesota, Minneapolis, MN) analyzed. Glucose dynamics were measured at fasting and during an oral glucose tolerance test (OGTT), using point of care instruments (DCA Analyzer, Siemens Medical Solutions, Malvern, PA; YSI Analyzers, Xylem Inc., Yellow Springs, OH). High added sugar intake was defined as consuming above the recommendation of 10% of calories from the US Dietary Guidelines. Independent sample T-tests assessed the differences between groups consuming high versus recommended amounts of added sugars. Values are expressed as mean ± standard deviation. Thirty-one adolescents, ages 15.5 ± 2.4 years, were screened. The sample was composed of 12 boys and 19 girls, and 45% had prediabetes. The BMI of the sample was 34.3 ± 6.8 kg/m2 with no differences between normal status and prediabetes groups. Similarly, normal status (11.2 ± 4.6%) and prediabetes (11.3 ± 5.0%) groups each consumed excess amounts of added sugars with no differences between groups. There were no significant differences between glycated hemoglobin (HbA1c, 5.5 ± 0.5% and 5.3 ± 0.2%), 2 hour glucose concentrations (125.4 ± 28.7 mg/dL and 111.9 ± 22.0 mg/dL), or BMI (33.9 ± 6.0 kg/m2 and 34.9 ± 8.2 kg/m2) between the groups with high versus recommended intakes of added sugar, respectively. The fasting plasma glucose concentrations in the group with high intakes of added sugar tended to be higher compared to the group with recommended intake of added sugar (94.6 ± 5.7 mg/dL versus 90.8 ± 5.1 mg/dL, P = 0.095). Fasting glucose may be higher in adolescents consuming excess compared to recommended amounts of added sugars. This research highlights the need for additional research to clarify the metabolic consequences of high amounts of added sugars in the diets of adolescents with obesity and a risk for developing type 2 diabetes. McKinley Foundation, Indiana CTSI Project Development Team UL1TR002529.
Recent research has shown a strong theoretical connection between variational autoencoders (VAEs) and the rate-distortion theory. Motivated by this, we consider the problem of lossy image compression from the perspective of generative modeling. Starting with ResNet VAEs, which are originally designed for data (image) distribution modeling, we redesign their latent variable model using a quantization-aware posterior and prior, enabling easy quantization and entropy coding at test time. Along with improved neural network architecture, we present a powerful and efficient model that outperforms previous methods on natural image lossy compression. Our model compresses images in a coarse-to-fine fashion and supports parallel encoding and decoding, leading to fast execution on GPUs. Code is available at https://github.com/duanzhiihao/lossy-vae.
'/%3e%3c/defs%3e%3cpath fill='%23012169' d='M0 0h10080v5040H0z'/%3e%3cpath stroke='%23fff' stroke-width='.6' d='m0 0 6 3m0-3L0 3' clip-path='url(%23b)' transform='scale(840)'/%3e%3cpath stroke='%23e4002b' stroke-width='.4' d='m0 0 6 3m0-3L0 3' clip-path='url(%23c)' transform='scale(840)'/%3e%3cpath stroke='%23fff' stroke-width='840' d='M2520 0v2520M0 1260h5040'/%3e%3cpath stroke='%23e4002b' stroke-width='504' d='M2520 0v2520M0 1260h5040'/%3e%3cg fill='%23fff'%3e%3cuse xlink:href='%23d' x='2520' y='3780'/%3e%3cuse xlink:href='%23a' x='7560' y='4200'/%3e%3cuse xlink:href='%23a' x='6300' y='2205'/%3e%3cuse xlink:href='%23a' x='7560' y='840'/%3e%3cuse xlink:href='%23a' x='8680' y='1869'/%3e%3cuse xlink:href='%23e' x='8064' y='2730'/%3e%3c/g%3e%3c/svg%3e)
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)