Image classification with small datasets has been an active research area in the recent past. However, as research in this scope is still in its infancy, two key ingredients are missing for ensuring ...reliable and truthful progress: a systematic and extensive overview of the state of the art, and a common benchmark to allow for objective comparisons between published methods. This article addresses both issues. First, we systematically organize and connect past studies to consolidate a community that is currently fragmented and scattered. Second, we propose a common benchmark that allows for an objective comparison of approaches. It consists of five datasets spanning various domains ( e.g. , natural images, medical imagery, satellite data) and data types (RGB, grayscale, multispectral). We use this benchmark to re-evaluate the standard cross-entropy baseline and ten existing methods published between 2017 and 2021 at renowned venues. Surprisingly, we find that thorough hyper-parameter tuning on held-out validation data results in a highly competitive baseline and highlights a stunted growth of performance over the years. Indeed, only a single specialized method dating back to 2019 clearly wins our benchmark and outperforms the baseline classifier.
A healthy diet can help to prevent or manage many important conditions and diseases, particularly obesity, malnutrition, and diabetes. Recent advancements in artificial intelligence and smartphone ...technologies have enabled applications to conduct automatic nutritional assessment from meal images, providing a convenient, efficient, and accurate method for continuous diet evaluation. We now extend the goFOODTM automatic system to perform food segmentation, recognition, volume, as well as calorie and macro-nutrient estimation from single images that are captured by a smartphone. In order to assess our system’s performance, we conducted a feasibility study with 50 participants from Switzerland. We recorded their meals for one day and then dietitians carried out a 24 h recall. We retrospectively analysed the collected images to assess the nutritional content of the meals. By comparing our results with the dietitians’ estimations, we demonstrated that the newly introduced system has comparable energy and macronutrient estimation performance with the previous method; however, it only requires a single image instead of two. The system can be applied in a real-life scenarios, and it can be easily used to assess dietary intake. This system could help individuals gain a better understanding of their dietary consumption. Additionally, it could serve as a valuable resource for dietitians, and could contribute to nutritional research.
This study explores if there is a convergence between the concepts of fashion and eco-friendliness in consumer perception of a fashion brand. We assume that increased eco-friendly perception will ...influence the brand image positively, with this impact being much higher for luxury than for high and fast fashion brands. The hypotheses are tested using data collected from Twitter. We analyzed the fashion clothing brands with the highest number of followers on the Socialbakers list and applied a novel social network mining methodology that allows measuring the relationship between each brand and two perceptual attributes (fashion and eco-friendliness). The method is based on attribute exemplars—that is, Twitter accounts that represent a perceptual attribute. Our exemplars catalyze social media conversations on fashion (identified in our research by the keywords “fashion,” “glamour,” and “style”) and eco-friendliness (keywords “environment” and “ethical business”). Based on social network analysis theory, we computed a similarity function between the followers of the exemplars and those of the brand. The results suggest that there is a correlation between the fashion and the eco-friendliness perceptual attributes of a brand; however, this correlation is far stronger for luxury brands than for high and fast fashion brands. The difference in the correlations confirms the recent tendency of fashion luxury brand to increasingly consider treating environmental issues as part of their core business and not just as added value to the brand’s offer.
•Eco-friendliness is becoming a fundamental component of the value proposition of fashion brands.•There is strong correlation between the fashion and the eco-friendliness perceptual attributes of luxury brands.•Consumers are more sensitive to fashion brands’ environmental rather than ethical business practices.
A healthy diet can help to prevent or manage many important conditions and diseases, particularly obesity, malnutrition, and diabetes. Recent advancements in artificial intelligence and smartphone ...technologies have enabled applications to conduct automatic nutritional assessment from meal images, providing a convenient, efficient, and accurate method for continuous diet evaluation. We now extend the goFOODsup.TM automatic system to perform food segmentation, recognition, volume, as well as calorie and macro-nutrient estimation from single images that are captured by a smartphone. In order to assess our system’s performance, we conducted a feasibility study with 50 participants from Switzerland. We recorded their meals for one day and then dietitians carried out a 24 h recall. We retrospectively analysed the collected images to assess the nutritional content of the meals. By comparing our results with the dietitians’ estimations, we demonstrated that the newly introduced system has comparable energy and macronutrient estimation performance with the previous method; however, it only requires a single image instead of two. The system can be applied in a real-life scenarios, and it can be easily used to assess dietary intake. This system could help individuals gain a better understanding of their dietary consumption. Additionally, it could serve as a valuable resource for dietitians, and could contribute to nutritional research.
We introduce Tune without Validation (Twin), a pipeline for tuning learning rate and weight decay without validation sets. We leverage a recent theoretical framework concerning learning phases in ...hypothesis space to devise a heuristic that predicts what hyper-parameter (HP) combinations yield better generalization. Twin performs a grid search of trials according to an early-/non-early-stopping scheduler and then segments the region that provides the best results in terms of training loss. Among these trials, the weight norm strongly correlates with predicting generalization. To assess the effectiveness of Twin, we run extensive experiments on 20 image classification datasets and train several families of deep networks, including convolutional, transformer, and feed-forward models. We demonstrate proper HP selection when training from scratch and fine-tuning, emphasizing small-sample scenarios.
Solving image classification tasks given small training datasets remains an open challenge for modern computer vision. Aggressive data augmentation and generative models are among the most ...straightforward approaches to overcoming the lack of data. However, the first fails to be agnostic to varying image domains, while the latter requires additional compute and careful design. In this work, we study alternative regularization strategies to push the limits of supervised learning on small image classification datasets. In particular, along with the model size and training schedule scaling, we employ a heuristic to select (semi) optimal learning rate and weight decay couples via the norm of model parameters. By training on only 1% of the original CIFAR-10 training set (i.e., 50 images per class) and testing on ciFAIR-10, a variant of the original CIFAR without duplicated images, we reach a test accuracy of 66.5%, on par with the best state-of-the-art methods.
Although remarkable improvements have been made, the natural control of hand prostheses in everyday life is still challenging. Changes in limb position can considerably affect the robustness of ...pattern recognition-based myoelectric control systems, even if various strategies were proposed to mitigate this effect. In this paper, we investigate the possibility of selecting a set of training movements that is robust to limb position change, performing a trade-off between training time and accuracy. Four able-bodied subjects were recorded while following a training protocol for myoelectric hand prostheses control. The protocol is composed of 210 combinations of arm positions, forearm orientations, wrist orientations and hand grasps. To the best of our knowledge, it is among the most complete including changes in limb positions. A training reduction paradigm was used to select subsets of training movements from a group of subjects that were tested on the left-out subject's data. The results show that a reduced training set (30 to 50 movements) allows a substantial reduction of the training time while maintaining reasonable performance, and that the trade-off between performance and training time appears to depend on the chosen classifier. Although further improvements can be made, the results show that properly selected training sets can be a viable strategy to reduce the training time while maximizing the performance of the classifier against variations in limb position.