Text-driven localized editing of 3D objects is particularly difficult as locally mixing the original object with intended new and style effects without distorting object's form not a straightforward process. To address this issue, we propose novel NeRF-based model, Blending-NeRF, which consists two NeRF networks: pre-trained editable NeRF. Additionally, introduce blending operations that allow Blending-NeRF to properly edit target regions are by text. By using pretrained vision-language...

The autonomous parking of vehicles requires the ability to accurately locate an available slot in vicinity a vehicle. Since slots have variety shapes and colors, may be occluded by obstacles, or look different due surroundings such as lighting, locating them can challenging task. In this paper, we propose context-based detection method inspired process human driver finding slot. Our consists two deep network modules: context recognizer detector. identifies environment (type, angle,...

This paper presents a blind digital video watermarking scheme, which is especially robust to camcorder recording attacks and also variety of common processing geometric distortions. Using the fact that nearby frames sequence are quite similar, method embeds watermark by temporal modulation frames. The pattern used in generated based on pixel-value histogram, makes extraction free from synchronization. To make it imperceptible, adjusted according roughly Human Visual System. experimental...

While 3D-based GAN techniques have been successfully applied to render photo-realistic 3D images with a variety of attributes while preserving view consistency, there has little research on how fine-control without limiting specific category objects their properties. To fill such gap, we propose novel image manipulation model representations for fine-grained control custom attributes. By extending the latest models (e.g., EG3D), our user-friendly quantitative enables fine yet normalized...

This paper presents a commercial implementation of CNN-based classification parking slot type using around view images. The existing automatic systems use ultrasonic sensors, but they often fail to classify the types slots. Around images can depict slots distinguishably. However, due diverse lighting and ground conditions, it is difficult Moreover, hard find lines since are occluded by vehicle or erased. To overcome these problems, we have constructed an extensive dataset composed labeled...

We present a new multi-modal face image generation method that converts text prompt and visual input, such as semantic mask or scribble map, into photo-realistic image. To do this, we combine the strengths of Generative Adversarial networks (GANs) diffusion models (DMs) by employing features in DM latent space pre-trained GANs. simple mapping style modulation network to link two convert meaningful representations feature maps attention codes. With GAN inversion, estimated codes can be used...

Interactive segmentation of 3D Gaussians opens a great opportunity for real-time manipulation scenes thanks to the rendering capability Gaussian Splatting. However, current methods suffer from time-consuming post-processing deal with noisy output. Also, they struggle provide detailed segmentation, which is important fine-grained scenes. In this study, we propose Click-Gaussian, learns distinguishable feature fields two-level granularity, facilitating without post-processing. We delve into...

Surround view monitoring (SVM) system provides a composite bird-eye of the vehicle to assist in safe parking. Since each camera independently performs auto exposure (AE) and white balance (AWB), has noticeable boundaries between adjacent views. To achieve seamlessly stitched view, we propose an effective photometric alignment for surround using simple additive gain model. Experimental results show that proposed method view. And processing time achieves 3ms on NVIDIA Tegra CX embedded platform.

Text-driven localized editing of 3D objects is particularly difficult as locally mixing the original object with intended new and style effects without distorting object's form not a straightforward process. To address this issue, we propose novel NeRF-based model, Blending-NeRF, which consists two NeRF networks: pretrained editable NeRF. Additionally, introduce blending operations that allow Blending-NeRF to properly edit target regions are by text. By using vision-language aligned CLIP,...