Diffusion models are rising as a powerful solution for high-fidelity image generation, which exceeds GANs in quality in many circumstances. However, their slow training and inference speed is a huge bottleneck, blocki...
详细信息
ISBN:
(纸本)9798350301298
Diffusion models are rising as a powerful solution for high-fidelity image generation, which exceeds GANs in quality in many circumstances. However, their slow training and inference speed is a huge bottleneck, blocking them from being used in real-time applications. A recent DiffusionGAN method significantly decreases the models' running time by reducing the number of sampling steps from thousands to several, but their speeds still largely lag behind the GAN counterparts. This paper aims to reduce the speed gap by proposing a novel wavelet-based diffusion scheme. We extract low-and-high frequency components from both image and feature levels via wavelet decomposition and adaptively handle these components for faster processing while maintaining good generation quality. Furthermore, we propose to use a reconstruction term, which effectively boosts the model training convergence. Experimental results on CelebA-HQ, CIFAR-10, LSUN-Church, and STL-10 datasets prove our solution is a stepping-stone to offering real-time and high-fidelity diffusion models. Our code and pre-trained checkpoints are available at https://***/VinAIResearch/***.
Multibeam echosounders (MBES) are the tool of choice for high-precision underwater surveys, especially when water conditions render optical imagery ineffective. We present and evaluate the following approaches for MBE...
详细信息
ISBN:
(纸本)9798350362077
Multibeam echosounders (MBES) are the tool of choice for high-precision underwater surveys, especially when water conditions render optical imagery ineffective. We present and evaluate the following approaches for MBES segmentation: (1) real-timeprocessing of single sounding profiles using traditional machine learning techniques, (2) batch processing of "waterfall" pseudo-images using a standard U-Net model, (3) the same model adapted to 2D projections of 3D point clouds, and (4) post-mission, survey-level processing using modern networks specifically designed for sparse point clouds. Strengths and weaknesses of the methods are discussed, including data preprocessing requirements, robustness, and ease of implementation/interpretation. Evaluation is performed on real data collected by an autonomous underwater vehicle (AUV) during a deep-sea industrial pipeline inspection.
In the field of medical imaging, C-arm systems play a pivotal role in surgeries, especially interventional surgeries. However, the current C-arm imaging system cannot adapt to the application scenarios due to algorith...
详细信息
In the real-timeimageprocessing system of the airborne infrared camera, how to process a large amount of data and information in a limited time and effectively meet the real-time requirements of the system is a prob...
详细信息
ISBN:
(数字)9781510652095
ISBN:
(纸本)9781510652095;9781510652088
In the real-timeimageprocessing system of the airborne infrared camera, how to process a large amount of data and information in a limited time and effectively meet the real-time requirements of the system is a problem that needs to be solved as soon as possible. Based on this, this article meets the needs of system modularization design ideas, and proposes a real-timeimageprocessing system design based on CMOS, using DSP and FPGA as the core devices to realize the corresponding hardware circuit design, and using high-demand tracking in the real-timeimageprocessing system algorithm. The experimental results show that the functions and performance of the designed real-timeimageprocessing system can meet the expected demand, and it is practical and reliable.
Garbage collection in urban areas has become a major challenge due to the increase in trash production. New technologies, including the application of deep learning and imageprocessing methods, have been created to s...
详细信息
images captured in low light conditions usually suffer from poor visibility, a high amount of noise, and little information stored in the dark image, which has a negative impact on subsequent processing for outdoor co...
详细信息
ISBN:
(纸本)9798350374292;9798350374285
images captured in low light conditions usually suffer from poor visibility, a high amount of noise, and little information stored in the dark image, which has a negative impact on subsequent processing for outdoor computer vision applications. Presently, numerous deep learning based methods achieved superior performance with multi-exposure paired training data or additional information. However, obtaining multi-exposure data samples is a tedious task in real-time scenarios. To mitigate this challenge, we propose a zero reference based learnable wavelet approach without multi-exposure paired training data requirement for low-light image enhancement. Our proposed approach generates the low light image and learns to project an image into noise free similar looking image, then we enhance the image using retinex theory. Further, we have proposed learnable wavelet block to remove the hidden noise amplified while enhancement. We introduce Gaussian-based supervision to improve the smoothness of the image. Extensive experimental analysis on synthetic as well as real-world images, along with thorough ablation study demonstrate the effectiveness of our proposed method over the existing state-of-the-art methods for low-light image enhancement. The code is provided at https://***/vision-lab-sggsiet/Zero-Reference-based-Low-light-Enhancement-with-Wavelet-Optimization.
In surveillance video, target tracking is an important part. Based on imageprocessing technology, this paper studies a real-time and effective method to collect and recognize camera motion information. Firstly, the i...
详细信息
ISBN:
(纸本)9798350310801
In surveillance video, target tracking is an important part. Based on imageprocessing technology, this paper studies a real-time and effective method to collect and recognize camera motion information. Firstly, the influence of visual dead angle and illumination on recognition is analyzed. Secondly, according to the characteristic of background light intensity, the corresponding algorithm is designed to realize the positioning and tracking control strategy of the target and surrounding environment scenery. Finally, the correctness of the method is verified by MATLAB simulation software, so as to obtain a better and scalable scheme, which is more economical and feasible after the occlusion rate is minimized.
In rainy scenarios, military target images captured by sensors are occluded by rain, leading to local information loss, which hampers the accurate reception and judgment of battlefield situation information. To tackle...
详细信息
In recent years, the integration of deep learning technologies in agriculture has shown significant potential for enhancing efficiency and productivity. This paper presents a novel fruit picking and quality analysis s...
详细信息
Fake image detection has become an urgent task. While advanced technology benefits us, it also poses a threat when used in cybercrime. images are often considered solid evidence to prove something concrete, and image ...
详细信息
暂无评论