Improved Semantic Representation Learning by Multiple Clustering for Image-Based 3D Model Retrieval

Abstract

Under the heavy management on the increasing 3D models, the topic of image-based 3D model retrieval which organizes unlabeled 3D models based on abundant knowledge learned from labeled 2D images has drawn attentions. However, prior methods are limited in aligning semantically at corresponding categories of two domains due to the lack of label information in 3D domain. To this end, this paper proposes an improved semantic representation learning by multiple clustering approach, which improves the reliability of pseudo labels for 3D models, so as to achieve class-level semantic alignment. Specifically, this paper first extracts features for 2D images and 3D models. Then it clusters combining the 3D features with the semantic information from multiple clustering on 3D model features to obtain more reliable target pseudo label. Extensive experiments have shown that the proposed method has achieved the gain of 3.0%-205.0% averagely for popular retrieval metrics on the benchmark of Monocular Image based 3D Object Retrieval (MI3DOR), and 1.3%-69.7% on another advanced benchmark MI3DOR-2.

Query-Guided Refinement and Dynamic Spans Network for Video Highlight Detection and Temporal Grounding in Online Information Systems

“…DRFT (Chen et al, 2021) integrates multi-modal features such as depth (Xu et al, 2021a;Xu et al, 2021b) and optical flow to learn complementary visual sources, but without a decoder it results in low accuracy of TG. 3D features (Chu et al, 2022;Srivastava et al, 2022) in the video also improve multi-modal representation, and GTR (Cao et al, 2021) utilizes a cubic embedding extractor to capture 3D features in videos. Yet 3D feature extracting is time-consuming.…”

A Semantic Feature Enhancement-Based Aerial Image Target Detection Method Using Dense RFB-FE

“…This study improved the SSD algorithm, as shown in Figure 1. In this study, a multi-scale ship target detection algorithm based on dense RFB and CGAM is proposed with the following main improvements: First, a dense RFB shallow feature enhancement module (dense RFB-FE) is designed to enhance the shallow detail information of the network; second, CGAM is designed to effectively extract the deep semantic information (Chu, et al, 2022); and third, we the loss function is designed, and the focused classification loss function is added.…”

“…Drones have been applied in various fields such as power detection, environmental protection, biological detection, logistics and transportation, disaster rescue, data collection, and mobile communication (Razakarivony., & Julie., 2016). In the coming years, the deep integration of drone technology with artificial intelligence (Li, D., et al, 2019;Nhi, et al, 2022), image processing (Chu, et al, 2022;Qian, et al, 2022;Zheng, et al, 2022), network security (Alomani, et al, 2022;Gaurav, et al, 2022) and other technologies will not only further overcome the problems of drones in current industrial production, it will also promote the landing of UAV applications in new fields (Betti., & Tucci., 2023;Ahmed, et al, 2022;Sun, et al, 2020;Zhang, et al, 2021). The wide application of drones in society has significantly improved production efficiency and also considerably reduced the consumption of human, material, and financial resources.…”

Enhanced Metaheuristics with Trust Aware Route Selection for Wireless Sensor Networks

“…Moreover, WSNs are more sensitive as the nodes are recurrently positioned in an uncomfortable range. Even though, several applications routes at the atmospheres which are unreliable that further desire a sheltered routing and transmission [8]. Fig.…”