Recently, differentiable search methods have made major progress in reducing the computational costs of neural architecture search. However, these approaches often report lower accuracy evaluating searched or transferring it to another dataset. This is arguably due large gap between depths and evaluation scenarios. In this paper, we present an efficient algorithm which allows depth architectures grow gradually during training procedure. brings two issues, namely, heavier overheads weaker...

With recent advances in mobile computing, the demand for visual localization or landmark identification on devices is gaining interest. We advance state of art this area by fusing two popular representations street-level image data - facade-aligned and viewpoint-aligned show that they contain complementary information can be exploited to significantly improve recall rates city scale. also feature detection low contrast parts data, discuss how incorporate priors a user's position (e.g. given...

Differentiable architecture search (DARTS) provided a fast solution in finding effective network architectures, but suffered from large memory and computing overheads jointly training super-network searching for an optimal architecture. In this paper, we present novel approach, namely, Partially-Connected DARTS, by sampling small part of to reduce the redundancy exploring space, thereby performing more efficient without comprising performance. particular, perform operation subset channels...

Electric vehicles emerge as the possible strategy for decarbonization and green transportation due to social demand. Researchers have made multiple efforts initiatives demand surge sustainable development in electric vehicle industry. This study analyzes relevant research of industry, thereby explores industry trends with a scientometrics-based data evaluation system, where three key topics are detected: "Vehicle Exhaust Emissions", "Climate Change", "Integration". The results visualized...

Recently, differentiable search methods have made major progress in reducing the computational costs of neural architecture search. However, these approaches often report lower accuracy evaluating searched or transferring it to another dataset. This is arguably due large gap between depths and evaluation scenarios. In this paper, we present an efficient algorithm which allows depth architectures grow gradually during training procedure. brings two issues, namely, heavier overheads weaker...

In recent years, the occurrence of rice pests has been increasing, which greatly affected yield in many parts world. The prevention and cure is urgent. Aiming at problems small appearance difference large size change various pests, a deep neural network named YOLO-GBS proposed this paper for detecting classifying from digital images. Based on YOLOv5s, one more detection head added to expand scale range, global context (GC) attention mechanism integrated find targets complex backgrounds,...

Differentiable architecture search (DARTS) enables effective neural (NAS) using gradient descent, but suffers from high memory and computational costs. In this paper, we propose a novel approach, namely Partially-Connected DARTS (PC-DARTS), to achieve efficient stable by reducing the channel spatial redundancies of super-network. level, partial connection is presented randomly sample small subset channels for operation selection accelerate process suppress over-fitting Side introduced...

In this paper, we study the problem of temporal video grounding (TVG), which aims to predict starting/ending time points moments described by a text sentence within long untrimmed video. Benefiting from fine-grained 3D visual features, TVG techniques have achieved remarkable progress in recent years. However, high complexity convolutional neural networks (CNNs) makes extracting dense features time-consuming, calls for intensive memory and computing resources. Towards efficient TVG, propose...

Deep Convolutional Neural Networks~(CNNs) offer remarkable performance of classifications and regressions in many high-dimensional problems have been widely utilized real-word cognitive applications. However, high computational cost CNNs greatly hinder their deployment resource-constrained applications, real-time systems edge computing platforms. To overcome this challenge, we propose a novel filter-pruning framework, two-phase filter pruning based on conditional entropy, namely...

Current 3D object detection paradigms highly rely on extensive annotation efforts, which makes them not practical in many real-world industrial applications. Inspired by that a human driver can keep accumulating experiences from self-exploring the roads without any tutor's guidance, we first step forwards to explore simple yet effective self-supervised learning framework tailored for LiDAR-based detection. Although pipeline has achieved great success 2D domain, characteristic challenges...

Instance segmentation of fruit tree canopies from images acquired by unmanned aerial vehicles (UAVs) is significance for the precise management orchards. Although deep learning methods have been widely used in fields feature extraction and classification, there are still phenomena complex data strong dependence on software performances. This paper proposes a learning-based instance method litchi trees, which has simple structure lower requirements form. Considering that models require large...

Arbuscular mycorrhizal fungi (AMF) colonize the rhizosphere of plants and form a symbiotic association with plants. Mycorrhizal symbionts have diversified ecological roles functions which are affected by soil conditions. Understanding effects different AMF inoculation on under varied nutritional conditions is great significance for further understanding external environment regulating symbiosis plant phenotypic traits. In this study, four treatments growth reproductive performance cherry...

Butterfly ecological image analysis (BEIA) is an exciting and essential field where computer vision can significantly aid in research biodiversity conservation. Although deep learning has made significant strides BEIA, the existing models still handle tasks such as segmentation classification independently, which constrains potential performance improvements gained by exploiting correlations between these tasks. In this paper, we design a multi-task model, named MdeBEIA, to perform both for...

Stereovision is the visual perception of depth derived from integration two slightly different images each eye, enabling understanding three-dimensional space. This capability deeply intertwined with cognitive brain functions. To explore impact stereograms varied motions on activities, we collected Electroencephalography (EEG) signals evoked by Dynamic Random Dot Stereograms (DRDS). effectively classify EEG induced DRDS, introduced a novel hybrid neural network model, XCF-LSTMSATNet, which...

The inferior reactive metal dispersion caused by lack of structural defects limits the application MCM-41 in synthesizing effective catalysts. Herein, we substituted conventional silicon source (CTAB) with attapulgite to...