đź“ť
Lu's Notes
  • Personal paper reading review
  • Data Mining
    • ABF-FNN: A new fuzzy neural network for predicting coal mine gas concentration hazard
    • FNDaaS: Content-agnostic Detection of Fake News sites
    • Robust Fraud Detection via Supervised Contrastive Learning
    • FNDaaS: Content-agnostic Detection of Fake News sites
  • Education
    • Combining Demographic Tabular Data with BERT Outputs for Multilabel Text Classification in Higher Ed
  • Environment
    • Predicting Impression Evaluation of Building Exterior Appearance Using Street Image Big Data and Dee
    • Combat Greenwashing with GoalSpotter: Automatic Sustainability Objective Detection in Heterogeneous
    • A Multi-task Model for Sentiment Aided Stance Detection of Climate Change Tweets
  • LLM
    • Can ChatGPT Reproduce Human-Generated Labels? A Study of Social Computing Tasks
    • What Can Large Language Models Tell Us about Time Series Analysis
  • Library Science
    • Analyzing academic mobility of U.S. professors based on ORCID data and the Carnegie Classificationtl
  • Security
    • How WEIRD is CHI
    • How WEIRD is Usable Privacy and Security Research?
    • Batman Hacked My Password: A Subtitle-Based Analysis of Password Depiction in Movies
    • Darknet
      • “It was honestly just gambling”: Investigating the Experiences of Teenage Cryptocurrency Users on Re
      • Does Online Anonymous Market Vendor Reputation Matter?
    • IllicitPromotion
      • Understanding Cross-Platform Referral Traffic for Illicit Drug Promotion
      • Seeing is Not Always Believing: An Empirical Analysis of Fake Evidence Generators
      • Unveiling the Risks of NFT Promotion Scams
      • Moderating Illicit Online Image Promotion for Unsafe User-Generated Content Games Using Large Vision
      • DarkGram: A Large-Scale Analysis of Cybercriminal Activity Channels on
      • Lifting the Grey Curtain: Analyzing the Ecosystem of Android Scam Apps
  • Social Media
    • Getting Meta A Multimodal Approach for Detecting Unsafe Conversations within Instagram Direct Messag
    • Sliding into My DMs: Detecting Uncomfortable or Unsafe Sexual Risk Experiences within Instagram Dire
    • Cross-Lingual and Cross-Domain Crisis Classification for Low-Resource Scenarios
    • On Xing Tian and the Perseverance of Anti-China Sentiment Online
    • Auditing Algorithmic Explanations of Social Media Feeds: A Case Study of TikTok Video Explanations
    • COVIDSenti: A Large-Scale Benchmark Twitter Data Set for COVID-19 Sentiment Analysis
    • “Go eat a bat, Chang!”: On the Emergence of Sinophobic Behavior on Web Communities in the Face of CO
    • TikTok and the Art of Personalization: Investigating Exploration and Exploitation on Social Media Fe
  • image
    • Title
  • Smart Aquaculture
    • Animal Re-identification
      • An Intelligence Cattle Reidentification System Over Transport by Siamese Neural Networks and YOLO
      • Convolutional Neural Networks for individual identification in the Southern Rock Lobster supply chai
      • Few-Shot Learning for Palmprint Recognition via Meta-Siamese Network
      • Re-identification of fish individuals of undulate skate via deep learning within a few-shot context
    • Computer Vision
      • Adapting Grad-CAM for Embedding Networks
      • CLIP-ReID: Exploiting Vision-Language Model for Image Re-Identificationwithout Concrete Text Labels
      • FastReID: A Pytorch Toolbox for General InstanceRe-identification
      • GW-net: An efficient grad-CAM consistency neural network with weakening of random erasing features f
      • Style Normalization and Restitution for Generalizable Person Re-Identification
    • Hyperspectral Imaging
      • Hyperspectral imaging–based assessment of fresh meat quality: Progress and applications
    • Sensor Data
      • A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anom
      • Big Data Driven Marine Environment Information Forecasting: A Time Series Prediction Network
      • Free nitrous acid prediction in ANAMMOX process using hybrid deep neural network model
      • Predicting ammonia nitrogen in surface water by a new attention-based deep learning hybrid model
      • Prediction of dissolved oxygen in a fishery pond based on gated recurrent unit (GRU)
      • Real-time in situ prediction of ocean currents
Powered by GitBook
On this page
  • Summary
  • Methodology
  • Emprical Study
  1. Environment

Predicting Impression Evaluation of Building Exterior Appearance Using Street Image Big Data and Dee

PreviousEnvironmentNextCombat Greenwashing with GoalSpotter: Automatic Sustainability Objective Detection in Heterogeneous

Last updated 2 months ago

Link: https://ieeexplore.ieee.org/document/10386316

Conference: IEEE Big Data 23

Keywords: Building Exterior Appearance, Street Image, Deep Learning

Summary

This paper proposes a method for predicting the impression evaluation of buildings’ exterior appearance using street image big data.

Research Question 1: Can the establishment of a method to create building exterior images linked to GIS data, including building location, shooting points, and attribute information, enable more detailed analyses in building exterior research using big image data? (GIS-based detailed analysis with image analysis)

Research Question 2: Previous studies have often analyzed impression evaluations for a single or a few attributes. However, when observing building exteriors, impressions are not necessarily singular but could be determined by interactions among multiple impressions. What kind of evaluative structure exists, and how are these impressions related to each other? (Correlations of different charactistics)

Research Question 3: While previous experiments have frequently compared images captured with the same composition or with limited building uses, differences in viewpoint angles or building uses might lead to different results even with the same experiment. In other words, could differences in building uses and viewpoint angles have some influence on the impression evaluations of building exteriors? (viewpoint angles and building uses' influence to impression evaluations)

Research Question 4: The predictive accuracy of models constructed in previous studies is good for test data, but the generalizability when inputting images beyond the training data remains unknown. Particularly, for models predicting subjective items, assessing the usefulness of the model is challenging. To what extent can the model constructed be applied to architectural design, especially considering its utility for subjective items? (generalizability)

Methodology

  1. Data Collection

  1. Building Selection Buildings that fit entirely within the omnidirectional street image.

  1. Creating Building Exterior Images

  1. Web Questionaire

10,000 pairs from the 1000 buildings. 16,722 participants. Each participant rate from 1-5 for 22 items.

  1. Prediction Model for Impression Evaluation

Siamese CNN. Input 2 images and predict the scores for 22 items.

Emprical Study

Omnidirectional Street Image: 360-degree street view images provided by a company

GIS Data: provided by the government

1000 buildings limited to 6 major uses.

Mask R-CNN for segmentation, then maunally extract

alt text
alt text
alt text