A5073577238- Advanced Neural Network Applications
- Human Pose and Action Recognition
- 3D Shape Modeling and Analysis
- Augmented Reality Applications
- Human Motion and Animation
- Face recognition and analysis
- Advanced Vision and Imaging
- CCD and CMOS Imaging Sensors
- Industrial Vision Systems and Defect Detection
- Generative Adversarial Networks and Image Synthesis
- COVID-19 diagnosis using AI
- AI in cancer detection
- Video Surveillance and Tracking Methods
- Computer Graphics and Visualization Techniques
- Advanced Image Fusion Techniques
- Digital Media Forensic Detection
- Image and Signal Denoising Methods
- Anomaly Detection Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Automated Road and Building Extraction
- Virtual Reality Applications and Impacts
- Face and Expression Recognition
- Biometric Identification and Security
- Neural Networks and Applications
- Advanced Image Processing Techniques
Samsung (United States)
2023-2024
Institute of Automation and Electrometry
2017-2022
Siberian Branch of the Russian Academy of Sciences
2022
Seoul National University
2021
Russian Academy of Sciences
2017
Motion blur is a common photography artifact in dynamic environments that typically comes jointly with the other types of degradation. This paper reviews NTIRE 2021 Challenge on Image Deblurring. In this challenge report, we describe specifics and evaluation results from 2 competition tracks proposed solutions. While both aim to recover high-quality clean image blurry image, different artifacts are involved. track 1, images low resolution while compressed JPEG format. each competition, there...
Computer vision has achieved impressive progress in recent years. Meanwhile, mobile phones have become the primary computing platforms for millions of people. In addition to phones, many autonomous systems rely on visual data making decisions, and some these limited energy (such as unmanned aerial vehicles also called drones robots). These batteries, efficiency is critical. This paper serves following two main purposes. First, examine state art low-power solutions detect objects images....
We present DINAR, an approach for creating realistic rigged fullbody avatars from single RGB images. Similarly to previous works, our method uses neural textures combined with the SMPL-X body model achieve photorealistic quality of while keeping them easy animate and fast infer. To restore texture, we use a latent diffusion show how such can be trained in texture space. The allows us realistically reconstruct large unseen regions as back person given frontal view. models pipeline are using...
We present a system to create Mobile Realistic Fullbody (MoRF) avatars. MoRF avatars are rendered in real-time on mobile devices, learned from monocular videos, and have high realism. use SMPL-X as proxy geometry render it with DNR (neural texture image-2-image network). improve prior work, by overfitting perframe warping fields the neural space, allowing better align training signal between different frames. also refine mesh fitting procedure overall avatar quality. In comparisons other...
We present HAHA - a novel approach for animatable human avatar generation from monocular input videos. The proposed method relies on learning the trade-off between use of Gaussian splatting and textured mesh efficient high fidelity rendering. demonstrate its efficiency to animate render full-body avatars controlled via SMPL-X parametric model. Our model learns apply only in areas where it is necessary, like hair out-of-mesh clothing. This results minimal number Gaussians being used represent...
Computer vision has achieved impressive progress in recent years. Meanwhile, mobile phones have become the primary computing platforms for millions of people. In addition to phones, many autonomous systems rely on visual data making decisions and some these limited energy (such as unmanned aerial vehicles also called drones robots). These batteries efficiency is critical. This article serves two main purposes: (1) Examine state-of-the-art low-power solutions detect objects images. Since...
The usage of convolutional neural networks (CNNs) in conjunction with a margin-based softmax approach demonstrates state-of-the-art performance for the face recognition problem. Recently, lightweight network models trained have been introduced identification task edge devices. In this paper, we propose novel distillation method architectures that outperforms other known methods on LFW, AgeDB-30 and Megaface datasets. idea proposed is to use class centers from teacher student network. Then...
We present billboard Splatting (BBSplat) - a novel approach for 3D scene representation based on textured geometric primitives. BBSplat represents the as set of optimizable planar primitives with learnable RGB textures and alpha-maps to control their shape. can be used in any Gaussian pipeline drop-in replacements Gaussians. Our method's qualitative quantitative improvements over 2D Gaussians are most noticeable when fewer used, achieves 1200 FPS. regularization term encourages have sparser...
The Low-Power Image Recognition Challenge (LPIRC, https://rebootingcomputing.ieee.org/lpirc) is an annual competition started in 2015. identifies the best technologies that can classify and detect objects images efficiently (short execution time low energy consumption) accurately (high precision). Over four years, winners' scores have improved more than 24 times. As computer vision widely used many battery-powered systems (such as drones mobile phones), need for low-power will become...
We present DINAR, an approach for creating realistic rigged fullbody avatars from single RGB images. Similarly to previous works, our method uses neural textures combined with the SMPL-X body model achieve photo-realistic quality of while keeping them easy animate and fast infer. To restore texture, we use a latent diffusion show how such can be trained in texture space. The allows us realistically reconstruct large unseen regions as back person given frontal view. models pipeline are using...
We present a system to create Mobile Realistic Fullbody (MoRF) avatars. MoRF avatars are rendered in real-time on mobile devices, learned from monocular videos, and have high realism. use SMPL-X as proxy geometry render it with DNR (neural texture image-2-image network). improve prior work, by overfitting per-frame warping fields the neural space, allowing better align training signal between different frames. also refine mesh fitting procedure overall avatar quality. In comparisons other...
Convolutional neural networks (CNN) allow achieving the highest accuracy for task of object detection in images. Major challenges further development detectors are false-positive detections and high demand processing power. In this paper, we propose an approach to which makes it possible reduce number by only moving objects required power algorithm inference. The proposed is a modification CNN already trained task. This method can be used improve existing system applying minor changes...