Deep reinforcement learning has proven to be a powerful approach to solving control tasks, but its characteristic high‑frequency oscillations make it difficult to apply in real‑world environments. While prior methods have addressed action oscillations via architectural or loss-based methods, the latter typically depend on heuristic or synthetic definitions of state similarity to promote action consistency, which often fail to accurately reflect the underlying system dynamics. In this paper, we propose a novel loss-based method by introducing a transition-induced similar state. The transition-induced similar state is defined as the distribution of next states transitioned from the previous state. Since it utilizes only environmental feedback and actually collected data, it better captures system dynamics. Building upon this foundation, we introduce Action Smoothing by Aligning Actions with Predictions from Preceding States (ASAP), an action smoothing method that effectively mitigates action oscillations. ASAP enforces action smoothness by aligning the actions with those taken in transition-induced similar states and by penalizing second-order differences to suppress high-frequency oscillations. Experiments in Gymnasium and Isaac-lab environments demonstrate that ASAP yields smoother control and improved policy performance over existing methods.

Visual Reinforcement Learning (RL) facilitates learning directly from raw images; however, the domain gap between training and testing environments frequently leads to a decline in performance within unseen environments. In this paper, we propose Fourier Guided Adaptive Adversarial Augmentation (FGA3), a novel augmentation method that maintains semantic consistency. We focus on style augmentation in the frequency domain by keeping the phase and altering the amplitude to preserve the state of the original data. For adaptive adversarial perturbation, we reformulate the worst-case problem to RL by employing adversarial example training, which leverages value loss and cosine similarity within a semantic space. Moreover, our findings illustrate that cosine similarity is effective in quantifying feature distances within a semantic space. Extensive experiments on DMControl-GB and Procgen have shown that FGA3 is compatible with a wide range of visual RL algorithms, both off-policy and on-policy, and significantly improves the robustness of the agent in unseen environments.

As concerns regarding privacy in deep learning continue to grow, individuals are increasingly apprehensive about the potential exploitation of their personal knowledge in trained models. Despite several research efforts to address this, they often fail to consider the real-world demand from users for complete knowledge erasure. Furthermore, our investigation reveals that existing methods have a risk of leaking personal knowledge through embedding features. To address these issues, we introduce a novel concept of Knowledge Deletion (KD), an advanced task that considers both concerns, and provides an appropriate metric, named Knowledge Retention score (KR), for assessing knowledge retention in feature space. To achieve this, we propose a novel training-free erasing approach named Erasing Space Concept (ESC), which restricts the important subspace for the forgetting knowledge by eliminating the relevant activations in the feature. In addition, we suggest ESC with Training (ESC-T), which uses a learnable mask to better balance the trade-off between forgetting and preserving knowledge in KD. Our extensive experiments on various datasets and models demonstrate that our proposed methods achieve the fastest and state-of-the-art performance. Notably, our methods are applicable to diverse forgetting scenarios, such as facial domain setting, demonstrating the generalizability of our methods. The code is available at https://github.com/KU-VGI/ESC.

Deep Neural Networks (DNNs) have become pivotal in various fields, especially in computer vision, outperforming previous methodologies. A critical challenge in their deployment is the bias inherent in data across different domains, such as image style and environmental conditions, leading to domain gaps. This necessitates techniques for learning general representations from biased training data, known as domain generalization. This paper presents Attend to eXpert Prompts (A2XP), a novel approach for domain generalization that preserves the privacy and integrity of the network architecture. A2XP consists of two phases: Expert Adaptation and Domain Generalization. In the first phase, prompts for each source domain are optimized to guide the model towards the optimal direction. In the second phase, two embedder networks are trained to effectively amalgamate these expert prompts, aiming for an optimal output. Our extensive experiments demonstrate that A2XP achieves state-of-the-art results over existing non-private domain generalization methods. The experimental results validate that the proposed approach not only tackles the domain generalization challenge in DNNs but also offers a privacy-preserving, efficient solution to the broader field of computer vision. Code is available at https://github.com/AIRLABkhu/A2XP.

Deep neural networks (DNNs) have successfully classified EEG-based brain-computer interface (BCI) systems. However, recent studies have found that well-designed input samples, known as adversarial examples, can easily fool well-performed deep neural networks model with minor perturbations undetectable by a human. This paper proposes an efficient generative model named generative perturbation network (GPN), which can generate universal adversarial examples with the same architecture for non-targeted and targeted attacks. Furthermore, the proposed model can be efficiently extended to conditionally or simultaneously generate perturbations for various targets and victim models. Our experimental evaluation demonstrates that perturbations generated by the proposed model outperform previous approaches for crafting signal-agnostic perturbations. We demonstrate that the extended network for signal-specific methods also significantly reduces generation time while performing similarly. The transferability across classification networks of the proposed method is superior to the other methods, which shows our perturbations' high level of generality.

Recent autostereoscopic 3-D (A3D) displays suffer from many limitations such as narrow viewing angle, low resolution, and shallow depth effects. As these limitations mainly originate from the insufficiency of pixel resources, it is not easy to obtain a feasible solution that can solve all the limitations simultaneously. In many cases, it will be better to find a good compromising design. Generally, the multiview display and the integral imaging display are the representative designs of A3D. However, as they are too canonical and lack flexibility in design, they tend to be a tradeoff. To address these design issues, we have analyzed the multiview display and the integral image display in a light-field coordinate and developed a 3-D display design framework in a light-field space. The developed framework does not use the “view” concept anymore. Instead, it considers the spatial distribution of rays of the 3-D display and provides more flexible and sophisticated design methods. In this paper, the developed design method is explained using a new pixel value assigning algorithm, called the light-field rendering, and vision-based parameter calibration methods for 3-D displays. We have also analyzed the blur effects caused by the depth and display characteristics. By implementing the proposed method, we have designed a 65-in 96-view display with a 4K panel. The developed prototype has showed almost seamless parallax with a high-resolution comparable to the conventional four to five views displays. This paper will be useful to readers interested in A3D displays, especially in the multiview and the integral imaging displays.

Nearly all 3D displays need calibration for correct rendering. More often than not, the optical elements in a 3D display are misaligned from the designed parameter setting. As a result, 3D magic does not perform well as intended. The observed images tend to get distorted. In this paper, we propose a novel display calibration method to fix the situation. In our method, a pattern image is displayed on the panel and a camera takes its pictures twice at different positions. Then, based on a quantitative model, we extract all display parameters (i.e., pitch, slanted angle, gap or thickness, and offset) from the observed patterns in the captured images. For high accuracy and robustness, our method analyzes the patterns mostly in the frequency domain. We conduct two types of experiments for validation; one with optical simulation for quantitative results and the other with real-life displays for qualitative assessment. Experimental results demonstrate that our method is quite accurate, about a half order of magnitude higher than prior work; is efficient, spending less than 2s for computation; and is robust to noise, working well in the SNR regime as low as 6dB.

The robust perception of robots is strongly needed to handle various objects skillfully. In this paper, we propose a novel approach to recognize objects and estimate their 6-DOF pose using 3D feature descriptors, called Geometric and Photometric Local Feature (GPLF). The proposed descriptors use both the geometric and photometric information of 3D point clouds from RGB-D camera and integrate those information into efficient descriptors. GPLF shows robust discriminative performance regardless of characteristics such as shapes or appearances of objects in cluttered scenes. The experimental results show how well the proposed approach classifies and identify objects. The performance of pose estimation is robust and stable enough for the robot to manipulate objects. We also compare the proposed approach with previous approaches that use partial information of objects with a representative large-scale RGB-D object dataset.

Feature extraction and matching is one of the most significant research areas in robot vision. In this paper, we present a new method for motion estimation and mapping using color feature extraction and matching. The proposed method reduces computational cost and has good performance. The experimental result shows that the proposed method not only runs faster but provides accurate result.