We propose Mask SSD, an efficient and effective approach to address the challenging instance segmentation task. Based on a single-shot detector, SSD detects all instances in image marks pixels that belong each instance. It consists of detection subnetwork predicts object categories bounding box locations, instance-level generates foreground mask for In subnetwork, multi-scale feedback features from different layers are used better represent objects various sizes provide high-level semantic...

Most successful object detectors are anchor-based, which is difficult to adapt the diversity of traffic objects. In this paper, we propose a novel anchor-free method, called FII-CenterNet, introduces foreground information eliminate interference complex background in scenes. The region proposal network segments based on boxes-induced segmentation annotation, and midground proposed provide rich edge addition location, scale also introduced improve regression performance. Extensive...

Power line parts detection refers to the inspection of key on transmission lines against complex background in aerial images and identifying whether exist anomalies that cause failure. Obviously, this process plays a pivotal role ensuring safety power transmission. Most existing methods are based deep convolutional neural networks. However, complexity variability image problem unmanned vehicles (UAVs) shooting perspective distance pose challenge for previous works. This study aims improve...

ABSTRACT Purpose: To explore the protective effects of borneol in myocardial ischemia-reperfusion injury (MIRI) and mechanism apoptosis. Methods: Cell viability was detected by CCK-8. The total superoxide dismutase (T-SOD) lactate dehydrogenase (LDH) leakage cells were tested biochemical assay kit. Detection apoptosis flow cytometry. Serum levels creatine kinase isoenzyme MB (CK-MB), LDH, cardiac troponin I (cTnI) enzyme-linked immunosorbent assay. Myocardial infarction area pathological...

Summary Light is a particularly important environmental cue that regulates variety of diverse plant developmental processes, such as photomorphogenesis. Blue light promotes photomorphogenesis mainly through the activation photoreceptor cryptochrome 1 (CRY1). However, mechanism underlying CRY1‐mediated regulation growth not fully understood. Here, we found blue induced N 6 ‐methyladenosine (m A) RNA modification during partially via CRY1. Cryptochrome mediates light‐induced expression...

Video Panoptic Segmentation (VPS) aims at assigning a class label to each pixel, uniquely segmenting and identifying all object instances consistently across frames. Classic solutions usually decompose the VPS task into several subtasks utilize multiple surrogates (e.g. boxes masks, centers offsets) represent objects. However, this divide-and-conquer strategy requires complex post-processing in both spatial temporal domains is vulnerable failures from surrogate tasks. In paper, inspired by...

This article aims to solve the problem of formation control mobile robots based on image and provide a low-cost as well ease-of-implementation solution for relying merely monocular camera under field-of-view (FOV) constraints. A low-complexity image-based visual servo controller is proposed, which can achieve desired relative position plane FOV constraints without feature depth leader's velocities information. To facilitate design, state transformation first performed decouple motion...

Object instance segmentation can achieve preferable results, powered with sufficient labeled training data. However, it is time-consuming for manually labeling, leading to the lack of large-scale diversified datasets accurate annotations. Exploiting synthetic data a very promising solution except domain distribution mismatch between dataset and real dataset. In this paper, we propose synthetic-to-real adaptation method object segmentation. At first, approach trained generate detection using...

Object detection has been widely adopted in video analysis and image understanding. Anchor-based object achieved good performance on the scale variation that is one long-standing problem for detection. The postprocessing an essential step of anchor-based after convolutional neural networks (CNN) it requires long computation time CPU or GPU. In this paper, we propose efficient FPGA solution using fixed-point representation postprocessing. quantization error mainly from sigmoid function...

The automatic identification of bird's nest in the inspection image transmission line is great significance to safe operation line. In this paper, a recognition method which combines visual saliency and depth learning proposed. This not only has advantage rich feature information visible light image, but also significant target. experimental results show that can accurately identify images with different background, tower shape, shooting angle distance, good robustness generalization,...

The perception of objects around the vehicle is important for both advanced driving assistance system (ADAS) and autonomous systems. However, it a huge challenge to recognize in low-light environment. In this paper, we propose monocular vision based nighttime driving. First, transnational videos are collected, which further processed into enhancement dataset object detection deep learning. Then, GAN-based EnlightenGAN trained enhancing image. CNN-based YOLOX inferred detect objects. Next,...

The key to 3D object detection is proper utilization of depth data. Compared with LiDAR based approaches, from a single image remains challenging task due the lack structure information. Recent methods leverage monocular estimation as way produce 2D maps, and adopt maps additional source input explore However, these either encode local correlations, or long range correlations by iteratively passing messages. In this work, we propose cross modal attention network (CMAN) for detection. It...

Image-level annotations allow to achieve semantic segmentation in a weakly-supervised way. Most advanced approaches utilize class activation map (CAM) from deep classifier generate pseudo-labels. However, CAM generally only focuses on the most discriminative parts of targets. To explore more pixel-level information and recognize all pixels within objects for segmentation, we propose Region-based Pixels Integration Mechanism (RPIM) which discovers intra-region inter-region information....

Automobile intelligence and networking have become the inevitable trend in future development of automotive industry. Existing intelligent connected vehicles rely on single-agent to perform basic perception, which is still weak dealing with problem accurate recognition positioning complex traffic scenes such as small far away objects. To tackle this issue, we propose a multi-model virtual-real fusion Transformer for collaborative perception. Specifically, possess complementary information...

Diffusion models have been recognized for their ability to generate images that are not only visually appealing but also of high artistic quality. As a result, Layout-to-Image (L2I) generation has proposed leverage region-specific positions and descriptions enable more precise controllable generation. However, previous methods primarily focus on UNet-based (e.g., SD1.5 SDXL), limited effort explored Multimodal Transformers (MM-DiTs), which demonstrated powerful image capabilities. Enabling...