Jong-Hwan Kim

Hwan-Soo Choi, Jongoh Jeong, Young Hoo Cho et al.

Sensor fusion approaches for intelligent self-driving agents remain key to driving scene understanding given visual global contexts acquired from input sensors. Specifically, for the local waypoint prediction task, single-modality networks are still limited by strong dependency on the sensitivity of the input sensor, and thus recent works therefore promote the use of multiple sensors in fusion in feature level in practice. While it is well known that multiple data modalities encourage mutual contextual exchange, it requires global 3D scene understanding in real-time with minimal computation upon deployment to practical driving scenarios, thereby placing greater significance on the training strategy given a limited number of practically usable sensors. In this light, we exploit carefully selected auxiliary tasks that are highly correlated with the target task of interest (e.g., traffic light recognition and semantic segmentation) by fusing auxiliary task features and also using auxiliary heads for waypoint prediction based on imitation learning. Our RGB-LIDAR-based multi-task feature fusion network, coined Cognitive TransFuser, augments and exceeds the baseline network by a significant margin for safer and more complete road navigation in the CARLA simulator. We validate the proposed network on the Town05 Short and Town05 Long Benchmark through extensive experiments, achieving up to 44.2 FPS real-time inference time.

Curie Kim, Ue-Hwan Kim, Jong-Hwan Kim

There have been attempts to detect 3D objects by fusion of stereo camera images and LiDAR sensor data or using LiDAR for pre-training and only monocular images for testing, but there have been less attempts to use only monocular image sequences due to low accuracy. In addition, when depth prediction using only monocular images, only scale-inconsistent depth can be predicted, which is the reason why researchers are reluctant to use monocular images alone. Therefore, we propose a method for predicting absolute depth and detecting 3D objects using only monocular image sequences by enabling end-to-end learning of detection networks and depth prediction networks. As a result, the proposed method surpasses other existing methods in performance on the KITTI 3D dataset. Even when monocular image and 3D LiDAR are used together during training in an attempt to improve performance, ours exhibit is the best performance compared to other methods using the same input. In addition, end-to-end learning not only improves depth prediction performance, but also enables absolute depth prediction, because our network utilizes the fact that the size of a 3D object such as a car is determined by the approximate size.

Hoyeon Lee, Hyun-Wook Yoon, Jong-Hwan Kim et al.

Phrase break prediction is a crucial task for improving the prosody naturalness of a text-to-speech (TTS) system. However, most proposed phrase break prediction models are monolingual, trained exclusively on a large amount of labeled data. In this paper, we address this issue for low-resource languages with limited labeled data using cross-lingual transfer. We investigate the effectiveness of zero-shot and few-shot cross-lingual transfer for phrase break prediction using a pre-trained multilingual language model. We use manually collected datasets in four Indo-European languages: one high-resource language and three with limited resources. Our findings demonstrate that cross-lingual transfer learning can be a particularly effective approach, especially in the few-shot setting, for improving performance in low-resource languages. This suggests that cross-lingual transfer can be inexpensive and effective for developing TTS front-end in resource-poor languages.

Tae-Min Choi, Jong-Hwan Kim

In this paper, we explore incremental few-shot object detection (iFSD), which incrementally learns novel classes using only a few examples without revisiting base classes. Previous iFSD works achieved the desired results by applying meta-learning. However, meta-learning approaches show insufficient performance that is difficult to apply to practical problems. In this light, we propose a simple fine-tuning-based approach, the Incremental Two-stage Fine-tuning Approach (iTFA) for iFSD, which contains three steps: 1) base training using abundant base classes with the class-agnostic box regressor, 2) separation of the RoI feature extractor and classifier into the base and novel class branches for preserving base knowledge, and 3) fine-tuning the novel branch using only a few novel class examples. We evaluate our iTFA on the real-world datasets PASCAL VOC, COCO, and LVIS. iTFA achieves competitive performance in COCO and shows a 30% higher AP accuracy than meta-learning methods in the LVIS dataset. Experimental results show the effectiveness and applicability of our proposed method.

In-Ug Yoon, Tae-Min Choi, Sun-Kyung Lee et al.

While many FSCIL studies have been undertaken, achieving satisfactory performance, especially during incremental sessions, has remained challenging. One prominent challenge is that the encoder, trained with an ample base session training set, often underperforms in incremental sessions. In this study, we introduce a novel training framework for FSCIL, capitalizing on the generalizability of the Contrastive Language-Image Pre-training (CLIP) model to unseen classes. We achieve this by formulating image-object-specific (IOS) classifiers for the input images. Here, an IOS classifier refers to one that targets specific attributes (like wings or wheels) of class objects rather than the image's background. To create these IOS classifiers, we encode a bias prompt into the classifiers using our specially designed module, which harnesses key-prompt pairs to pinpoint the IOS features of classes in each session. From an FSCIL standpoint, our framework is structured to retain previous knowledge and swiftly adapt to new sessions without forgetting or overfitting. This considers the updatability of modules in each session and some tricks empirically found for fast convergence. Our approach consistently demonstrates superior performance compared to state-of-the-art methods across the miniImageNet, CIFAR100, and CUB200 datasets. Further, we provide additional experiments to validate our learned model's ability to achieve IOS classifiers. We also conduct ablation studies to analyze the impact of each module within the architecture.

Curie Kim, Yewon Hwang, Jong-Hwan Kim

Reinforcement learning has shown an outstanding performance in the applications of games, particularly in Atari games as well as Go. Based on these successful examples, we attempt to apply one of the well-known reinforcement learning algorithms, Deep Q-Network, to the AI Soccer game. AI Soccer is a 5:5 robot soccer game where each participant develops an algorithm that controls five robots in a team to defeat the opponent participant. Deep Q-Network is designed to implement our original rewards, the state space, and the action space to train each agent so that it can take proper actions in different situations during the game. Our algorithm was able to successfully train the agents, and its performance was preliminarily proven through the mini-competition against 10 teams wishing to take part in the AI Soccer international competition. The competition was organized by the AI World Cup committee, in conjunction with the WCG 2019 Xi'an AI Masters. With our algorithm, we got the achievement of advancing to the round of 16 in this international competition with 130 teams from 39 countries.

Jongoh Jeong, Taek-Jin Song, Jong-Hwan Kim et al.

Understanding road scenes for visual perception remains crucial for intelligent self-driving cars. In particular, it is desirable to detect unexpected small road hazards reliably in real-time, especially under varying adverse conditions (e.g., weather and daylight). However, existing road driving datasets provide large-scale images acquired in either normal or adverse scenarios only, and often do not contain the road obstacles captured in the same visual domain as for the other classes. To address this, we introduce a new dataset called AVOID, the Adverse Visual Conditions Dataset, for real-time obstacle detection collected in a simulated environment. AVOID consists of a large set of unexpected road obstacles located along each path captured under various weather and time conditions. Each image is coupled with the corresponding semantic and depth maps, raw and semantic LiDAR data, and waypoints, thereby supporting most visual perception tasks. We benchmark the results on high-performing real-time networks for the obstacle detection task, and also propose and conduct ablation studies using a comprehensive multi-task network for semantic segmentation, depth and waypoint prediction tasks.

Ue-Hwan Kim, Yewon Hwang, Sun-Kyung Lee et al.

In this paper, we introduce a new benchmark dataset for the challenging writing in the air (WiTA) task -- an elaborate task bridging vision and NLP. WiTA implements an intuitive and natural writing method with finger movement for human-computer interaction (HCI). Our WiTA dataset will facilitate the development of data-driven WiTA systems which thus far have displayed unsatisfactory performance -- due to lack of dataset as well as traditional statistical models they have adopted. Our dataset consists of five sub-datasets in two languages (Korean and English) and amounts to 209,926 video instances from 122 participants. We capture finger movement for WiTA with RGB cameras to ensure wide accessibility and cost-efficiency. Next, we propose spatio-temporal residual network architectures inspired by 3D ResNet. These models perform unconstrained text recognition from finger movement, guarantee a real-time operation by processing 435 and 697 decoding frames-per-second for Korean and English, respectively, and will serve as an evaluation standard. Our dataset and the source codes are available at https://github.com/Uehwan/WiTA.

Joonhyuk Kim, Sahng-Min Yoo, Gyeong-Moon Park et al.

Unsupervised Domain Adaptation (UDA) for semantic segmentation has been favorably applied to real-world scenarios in which pixel-level labels are hard to be obtained. In most of the existing UDA methods, all target data are assumed to be introduced simultaneously. Yet, the data are usually presented sequentially in the real world. Moreover, Continual UDA, which deals with more practical scenarios with multiple target domains in the continual learning setting, has not been actively explored. In this light, we propose Continual UDA for semantic segmentation based on a newly designed Expanding Target-specific Memory (ETM) framework. Our novel ETM framework contains Target-specific Memory (TM) for each target domain to alleviate catastrophic forgetting. Furthermore, a proposed Double Hinge Adversarial (DHA) loss leads the network to produce better UDA performance overall. Our design of the TM and training objectives let the semantic segmentation network adapt to the current target domain while preserving the knowledge learned on previous target domains. The model with the proposed framework outperforms other state-of-the-art models in continual learning settings on standard benchmarks such as GTA5, SYNTHIA, CityScapes, IDD, and Cross-City datasets. The source code is available at https://github.com/joonh-kim/ETM.

Ue-Hwan Kim, Dongho Ka, Hwasoo Yeo et al.

Minimizing traffic accidents between vehicles and pedestrians is one of the primary research goals in intelligent transportation systems. To achieve the goal, pedestrian orientation recognition and prediction of pedestrian's crossing or not-crossing intention play a central role. Contemporary approaches do not guarantee satisfactory performance due to limited field-of-view, lack of generalization, and high computational complexity. To overcome these limitations, we propose a real-time predictive pedestrian collision warning service (P2CWS) for two tasks: pedestrian orientation recognition (100.53 FPS) and intention prediction (35.76 FPS). Our framework obtains satisfying generalization over multiple sites because of the proposed site-independent features. At the center of the feature extraction lies 3D pose estimation. The 3D pose analysis enables robust and accurate recognition of pedestrian orientations and prediction of intentions over multiple sites. The proposed vision framework realizes 89.3% accuracy in the behavior recognition task on the TUD dataset without any training process and 91.28% accuracy in intention prediction on our dataset achieving new state-of-the-art performance. To contribute to the corresponding research community, we make our source codes public which are available at https://github.com/Uehwan/VisionForPedestrian

Jongoh Jeong, Jong-Hwan Kim

Road scene understanding tasks have recently become crucial for self-driving vehicles. In particular, real-time semantic segmentation is indispensable for intelligent self-driving agents to recognize roadside objects in the driving area. As prior research works have primarily sought to improve the segmentation performance with computationally heavy operations, they require far significant hardware resources for both training and deployment, and thus are not suitable for real-time applications. As such, we propose a doubly contrastive approach to improve the performance of a more practical lightweight model for self-driving, specifically under adverse weather conditions such as fog, nighttime, rain and snow. Our proposed approach exploits both image- and pixel-level contrasts in an end-to-end supervised learning scheme without requiring a memory bank for global consistency or the pretraining step used in conventional contrastive methods. We validate the effectiveness of our method using SwiftNet on the ACDC dataset, where it achieves up to 1.34%p improvement in mIoU (ResNet-18 backbone) at 66.7 FPS (2048x1024 resolution) on a single RTX 3080 Mobile GPU at inference. Furthermore, we demonstrate that replacing image-level supervision with self-supervision achieves comparable performance when pre-trained with clear weather images.

Hoyeon Lee, Sejung Son, Ye-Eun Kang et al.

Current approaches to phrase break prediction address crucial prosodic aspects of text-to-speech systems but heavily rely on vast human annotations from audio or text, incurring significant manual effort and cost. Inherent variability in the speech domain, driven by phonetic factors, further complicates acquiring consistent, high-quality data. Recently, large language models (LLMs) have shown success in addressing data challenges in NLP by generating tailored synthetic data while reducing manual annotation needs. Motivated by this, we explore leveraging LLM to generate synthetic phrase break annotations, addressing the challenges of both manual annotation and speech-related tasks by comparing with traditional annotations and assessing effectiveness across multiple languages. Our findings suggest that LLM-based synthetic data generation effectively mitigates data challenges in phrase break prediction and highlights the potential of LLMs as a viable solution for the speech domain.

In-Ug Yoon, Tae-Min Choi, Young-Min Kim et al.

Few-shot class-incremental learning (FSCIL) presents the primary challenge of balancing underfitting to a new session's task and forgetting the tasks from previous sessions. To address this challenge, we develop a simple yet powerful learning scheme that integrates effective methods for each core component of the FSCIL network, including the feature extractor, base session classifiers, and incremental session classifiers. In feature extractor training, our goal is to obtain balanced generic representations that benefit both current viewable and unseen or past classes. To achieve this, we propose a balanced supervised contrastive loss that effectively balances these two objectives. In terms of classifiers, we analyze and emphasize the importance of unifying initialization methods for both the base and incremental session classifiers. Our method demonstrates outstanding ability for new task learning and preventing forgetting on CUB200, CIFAR100, and miniImagenet datasets, with significant improvements over previous state-of-the-art methods across diverse metrics. We conduct experiments to analyze the significance and rationale behind our approach and visualize the effectiveness of our representations on new tasks. Furthermore, we conduct diverse ablation studies to analyze the effects of each module.

Sahng-Min Yoo, Ue-Hwan Kim, Yewon Hwang et al.

Contemporary soft keyboards possess limitations: the lack of physical feedback results in an increase of typos, and the interface of soft keyboards degrades the utility of the screen. To overcome these limitations, we propose an Invisible Mobile Keyboard (IMK), which lets users freely type on the desired area without any constraints. To facilitate a data-driven IMK decoding task, we have collected the most extensive text-entry dataset (approximately 2M pairs of typing positions and the corresponding characters). Additionally, we propose our baseline decoder along with a semantic typo correction mechanism based on self-attention, which decodes such unconstrained inputs with high accuracy (96.0%). Moreover, the user study reveals that the users could type faster and feel convenience and satisfaction to IMK with our decoder. Lastly, we make the source code and the dataset public to contribute to the research community.

Dong He, Jie Cheng, Jong-Hwan Kim

Ground segmentation of point clouds remains challenging because of the sparse and unordered data structure. This paper proposes the GSECnet - Ground Segmentation network for Edge Computing, an efficient ground segmentation framework of point clouds specifically designed to be deployable on a low-power edge computing unit. First, raw point clouds are converted into a discretization representation by pillarization. Afterward, features of points within pillars are fed into PointNet to get the corresponding pillars feature map. Then, a depthwise-separable U-Net with the attention module learns the classification from the pillars feature map with an enormously diminished model parameter size. Our proposed framework is evaluated on SemanticKITTI against both point-based and discretization-based state-of-the-art learning approaches, and achieves an excellent balance between high accuracy and low computing complexity. Remarkably, our framework achieves the inference runtime of 135.2 Hz on a desktop platform. Moreover, experiments verify that it is deployable on a low-power edge computing unit powered 10 watts only.

Ue-Hwan Kim, Jong-Hwan Kim

Self-supervised learning of depth map prediction and motion estimation from monocular video sequences is of vital importance -- since it realizes a broad range of tasks in robotics and autonomous vehicles. A large number of research efforts have enhanced the performance by tackling illumination variation, occlusions, and dynamic objects, to name a few. However, each of those efforts targets individual goals and endures as separate works. Moreover, most of previous works have adopted the same CNN architecture, not reaping architectural benefits. Therefore, the need to investigate the inter-dependency of the previous methods and the effect of architectural factors remains. To achieve these objectives, we revisit numerous previously proposed self-supervised methods for joint learning of depth and motion, perform a comprehensive empirical study, and unveil multiple crucial insights. Furthermore, we remarkably enhance the performance as a result of our study -- outperforming previous state-of-the-art performance.

Jin-Man Park, Jae-Hyuk Jang, Sahng-Min Yoo et al.

We present a challenging dataset, ChangeSim, aimed at online scene change detection (SCD) and more. The data is collected in photo-realistic simulation environments with the presence of environmental non-targeted variations, such as air turbidity and light condition changes, as well as targeted object changes in industrial indoor environments. By collecting data in simulations, multi-modal sensor data and precise ground truth labels are obtainable such as the RGB image, depth image, semantic segmentation, change segmentation, camera poses, and 3D reconstructions. While the previous online SCD datasets evaluate models given well-aligned image pairs, ChangeSim also provides raw unpaired sequences that present an opportunity to develop an online SCD model in an end-to-end manner, considering both pairing and detection. Experiments show that even the latest pair-based SCD models suffer from the bottleneck of the pairing process, and it gets worse when the environment contains the non-targeted variations. Our dataset is available at http://sammica.github.io/ChangeSim/.

Tae-Min Choi, Ji-Su Kang, Jong-Hwan Kim

Time-series data with missing values are commonly encountered in many fields, such as healthcare, meteorology, and robotics. The imputation aims to fill the missing values with valid values. Most imputation methods trained the models implicitly because missing values have no ground truth. In this paper, we propose Random Drop Imputation with Self-training (RDIS), a novel training method for time-series data imputation models. In RDIS, we generate extra missing values by applying a random drop on the observed values in incomplete data. We can explicitly train the imputation models by filling in the randomly dropped values. In addition, we adopt self-training with pseudo values to exploit the original missing values. To improve the quality of pseudo values, we set the threshold and filter them by calculating the entropy. To verify the effectiveness of RDIS on the time series imputation, we test RDIS to various imputation models and achieve competitive results on two real-world datasets.

Ue-Hwan Kim, Se-Ho Kim, Jong-Hwan Kim

Intelligent agents need to understand the surrounding environment to provide meaningful services to or interact intelligently with humans. The agents should perceive geometric features as well as semantic entities inherent in the environment. Contemporary methods in general provide one type of information regarding the environment at a time, making it difficult to conduct high-level tasks. Moreover, running two types of methods and associating two resultant information requires a lot of computation and complicates the software architecture. To overcome these limitations, we propose a neural architecture that simultaneously performs both geometric and semantic tasks in a single thread: simultaneous visual odometry, object detection, and instance segmentation (SimVODIS). Training SimVODIS requires unlabeled video sequences and the photometric consistency between input image frames generates self-supervision signals. The performance of SimVODIS outperforms or matches the state-of-the-art performance in pose estimation, depth map prediction, object detection, and instance segmentation tasks while completing all the tasks in a single thread. We expect SimVODIS would enhance the autonomy of intelligent agents and let the agents provide effective services to humans.

Yong-Ho Yoo, Ue-Hwan Kim, Jong-Hwan Kim

Surface mount technology (SMT) is a process for producing printed circuit boards. Solder paste printer (SPP), package mounter, and solder reflow oven are used for SMT. The board on which the solder paste is deposited from the SPP is monitored by solder paste inspector (SPI). If SPP malfunctions due to the printer defects, the SPP produces defective products, and then abnormal patterns are detected by SPI. In this paper, we propose a convolutional recurrent reconstructive network (CRRN), which decomposes the anomaly patterns generated by the printer defects, from SPI data. CRRN learns only normal data and detects anomaly pattern through reconstruction error. CRRN consists of a spatial encoder (S-Encoder), a spatiotemporal encoder and decoder (ST-Encoder-Decoder), and a spatial decoder (S-Decoder). The ST-Encoder-Decoder consists of multiple convolutional spatiotemporal memories (CSTMs) with ST-Attention mechanism. CSTM is developed to extract spatiotemporal patterns efficiently. Additionally, a spatiotemporal attention (ST-Attention) mechanism is designed to facilitate transmitting information from the ST-Encoder to the ST-Decoder, which can solve the long-term dependency problem. We demonstrate the proposed CRRN outperforms the other conventional models in anomaly detection. Moreover, we show the discriminative power of the anomaly map decomposed by the proposed CRRN through the printer defect classification.

Ue-Hwan Kim, Jin-Man Park, Taek-Jin Song et al.

Intelligent agents gather information and perceive semantics within the environments before taking on given tasks. The agents store the collected information in the form of environment models that compactly represent the surrounding environments. The agents, however, can only conduct limited tasks without an efficient and effective environment model. Thus, such an environment model takes a crucial role for the autonomy systems of intelligent agents. We claim the following characteristics for a versatile environment model: accuracy, applicability, usability, and scalability. Although a number of researchers have attempted to develop such models that represent environments precisely to a certain degree, they lack broad applicability, intuitive usability, and satisfactory scalability. To tackle these limitations, we propose 3-D scene graph as an environment model and the 3-D scene graph construction framework. The concise and widely used graph structure readily guarantees usability as well as scalability for 3-D scene graph. We demonstrate the accuracy and applicability of the 3-D scene graph by exhibiting the deployment of the 3-D scene graph in practical applications. Moreover, we verify the performance of the proposed 3-D scene graph and the framework by conducting a series of comprehensive experiments under various conditions.

Ue-Hwan Kim, Sahng-Min Yoo, Jong-Hwan Kim

Text-entry aims to provide an effective and efficient pathway for humans to deliver their messages to computers. With the advent of mobile computing, the recent focus of text-entry research has moved from physical keyboards to soft keyboards. Current soft keyboards, however, increase the typo rate due to lack of tactile feedback and degrade the usability of mobile devices due to their large portion on screens. To tackle these limitations, we propose a fully imaginary keyboard (I-Keyboard) with a deep neural decoder (DND). The invisibility of I-Keyboard maximizes the usability of mobile devices and DND empowered by a deep neural architecture allows users to start typing from any position on the touch screens at any angle. To the best of our knowledge, the eyes-free ten-finger typing scenario of I-Keyboard which does not necessitate both a calibration step and a predefined region for typing is first explored in this work. For the purpose of training DND, we collected the largest user data in the process of developing I-Keyboard. We verified the performance of the proposed I-Keyboard and DND by conducting a series of comprehensive simulations and experiments under various conditions. I-Keyboard showed 18.95% and 4.06% increases in typing speed (45.57 WPM) and accuracy (95.84%), respectively over the baseline.

Ue-Hwan Kim, Jong-Hwan Kim

The automated home referred to as Smart Home is expected to offer fully customized services to its residents, reducing the amount of home labor, thus improving human beings' welfare. Service robots and Internet of Things (IoT) play the key roles in the development of Smart Home. The service provision with these two main components in a Smart Home environment requires: 1) learning and reasoning algorithms and 2) the integration of robot and IoT systems. Conventional computational intelligence-based learning and reasoning algorithms do not successfully manage dynamic changes in the Smart Home data, and the simple integrations fail to fully draw the synergies from the collaboration of the two systems. To tackle these limitations, we propose: 1) a stabilized memory network with a feedback mechanism which can learn user behaviors in an incremental manner and 2) a robot-IoT service provision framework for a Smart Home which utilizes the proposed memory architecture as a learning and reasoning module and exploits synergies between the robot and IoT systems. We conduct a set of comprehensive experiments under various conditions to verify the performance of the proposed memory architecture and the service provision framework and analyze the experiment results.