Search

While adversarial perturbation of images to attack deep image classification models pose serious security concerns in practice, this paper suggests a novel paradigm where the concept of image perturbation can benefit classification performance, which we call amicable aid. We show that by taking the opposite search direction of perturbation, an image can be modified to yield higher classification confidence and even a misclassified image can be made correctly classified. This can be also achieved with a large amount of perturbation by which the image is made unrecognizable by human eyes. The mechanism of the amicable aid is explained in the viewpoint of the underlying natural image manifold. Furthermore, we investigate the universal amicable aid, i.e., a fixed perturbation can be applied to multiple images to improve their classification results. While it is challenging to find such perturbations, we show that making the decision boundary as perpendicular to the image manifold as possible via training with modified data is effective to obtain a model for which universal amicable perturbations are more easily found., Comment: 5 pages

Recently, the vulnerability of deep image classification models to adversarial attacks has been investigated. However, such an issue has not been thoroughly studied for image-to-image tasks that take an input image and generate an output image (e.g., colorization, denoising, deblurring, etc.) This paper presents comprehensive investigations into the vulnerability of deep image-to-image models to adversarial attacks. For five popular image-to-image tasks, 16 deep models are analyzed from various standpoints such as output quality degradation due to attacks, transferability of adversarial examples across different tasks, and characteristics of perturbations. We show that unlike image classification tasks, the performance degradation on image-to-image tasks largely differs depending on various factors, e.g., attack methods and task objectives. In addition, we analyze the effectiveness of conventional defense methods used for classification models in improving the robustness of the image-to-image models., Comment: ICPR2022

Background Mandibular fractures are one of the most common types of facial fractures, the treatment of which can be delayed due to the severity of the trauma resulting in an increase of complications; thus, early evaluation of trauma severity at the time of visit is important. In South Korea, trauma patients are triaged and intensively treated in designated regional trauma centers. This study aimed to analyze the relationship between trauma severity and mandibular fracture patterns. Methods A medical records review was performed on patients who visited the regional trauma center at our hospital for mandibular fracture between 2009 and 2018. Epidemiologic data and mandibular fracture patterns were analyzed and compared with the conventional facial injury severity scale (FISS). Results Among 73 patients, 51 were classified as non-severe trauma patients and 22 as severe trauma patients. A higher trauma severity was associated with older age (odds ratio [OR], 1.164; 95% confidence interval [CI], 1.057–1.404) and lower risk was associated with fractures located in the angle (OR, 0.001; 95% CI, 0–0.022), condylar process (OR, 0.001; 95% CI, 0–0.28), and coronoid process (OR, 0.004; 95% CI, 0–0.985). The risk was lower when the injury mechanism was a pedestrian traffic accident (OR, 0.004; 95% CI, 0–0.417) or fall (OR, 0.004; 95% CI, 0–0.663) compared with an in-car traffic accident. Higher FISS (OR, 1.503; 95% CI, 1.155–2.049) was associated with a higher trauma severity. The proposed model was found to predict the trauma severity better than the model using FISS (p< 0.001). Conclusion Age, location of mandibular fractures, and injury mechanism showed significant relationships with the trauma severity. Epidemiologic data and patterns of mandibular fractures could predict the trauma severity better than FISS.

Additional file 1. The number of tsetse flies harboring 12S ribosomal RNA genes from several animals.

The video-based action recognition task has been extensively studied in recent years. In this paper, we study the structural vulnerability of deep learning-based action recognition models against the adversarial attack using the one frame attack that adds an inconspicuous perturbation to only a single frame of a given video clip. Our analysis shows that the models are highly vulnerable against the one frame attack due to their structural properties. Experiments demonstrate high fooling rates and inconspicuous characteristics of the attack. Furthermore, we show that strong universal one frame perturbations can be obtained under various scenarios. Our work raises the serious issue of adversarial vulnerability of the state-of-the-art action recognition models in various perspectives., Accepted by ICCV 2021

Recently, video-based action recognition methods using convolutional neural networks (CNNs) achieve remarkable recognition performance. However, there is still lack of understanding about the generalization mechanism of action recognition models. In this paper, we suggest that action recognition models rely on the motion information less than expected, and thus they are robust to randomization of frame orders. Based on this observation, we develop a novel defense method using temporal shuffling of input videos against adversarial attacks for action recognition models. Another observation enabling our defense method is that adversarial perturbations on videos are sensitive to temporal destruction. To the best of our knowledge, this is the first attempt to design a defense method specific to video-based action recognition models.

This paper reviews the AIM 2020 challenge on efficient single image super-resolution with focus on the proposed solutions and results. The challenge task was to super-resolve an input image with a magnification factor x4 based on a set of prior examples of low and corresponding high resolution images. The goal is to devise a network that reduces one or several aspects such as runtime, parameter count, FLOPs, activations, and memory consumption while at least maintaining PSNR of MSRResNet. The track had 150 registered participants, and 25 teams submitted the final results. They gauge the state-of-the-art in efficient single image super-resolution.

International audience

The placement of Korean fire-fighting administrative power in urban areas shows regional unbalance. In an ideal system, all citizens would be provided with equal fire protection and rescue services, but this is usually difficult to realize due to regional conditions or budget problems. In the case of Busan Metropolitan City, we deduced that it is impossible for half of the buildings to receive rescue services within 5 min, and the conditions are much worse for areas with long or wide fire-fighting service regions. The approach vulnerability for the existing emergency rescue squad locations was assessed. The results revealed that if a rescue team's location is shifted, the improvement effect will be virtually insignificant because of their geographical position. Therefore, this study suggests the establishment of additional rescue squads. It is proven that adding 5 rescue centers in the following locations could solve the problem of approach vulnerability: Bukbu, Gangseo, Geumjeong, Gijang, and Haeundae, in order of effectiveness. The number of buildings in the areas is 53,546.

Whole device printing is presented for realizing full colour displays with red (R), green (G) and blue (B) organic light emitting diodes (OLEDs). In this process, the whole OLED structure is transferred from a patterned mould to a glass substrate. Therefore, a simple step and repeat of the transfer of each of R, G and B OLED for RGB pixels completes the fabrication of the full colour display over a given area. A difference in the work of adhesion at two interfaces enables the transfer. A 'rigiflex' mould is used for the printing. It is rigid enough to allow sub-100 nm resolution and yet flexible enough for intimate contact with the glass substrate, which permits large area application.

Recently, evacuation analysis, which is one of the performance-based evacuation design methods, has been widely used to estimate the egress capacity of buildings and analyze evacuees’ characteristics. But almost all the models do not consider accurate evacuation scenarios because of the limitation of simulations, especially when setting up the staircase-length in building geometry mode. Therefore, this study selected a subject of a high-rise building with 351 invited participants, and conducted an experiment to investigate the evacuation from the 30th floor to the ground level and then classified overall 5,265 movement cases to calculate the participants' evacuation route. After the trial experiment, the paper calculated the precise average distance of participants' evacuation route in staircases considering each zone’s length and dimension and derived the numerical equations for inputting the parameter of staircase-length in evacuation models. And the verifications were also conducted by comparing simulations with the results of the experiment.