Portfolio
Computational Design Selected Works 2017-2021
Wenjie Huang Computer Science
Wenjie Huang Tel : +86 13719192209 Email : JankinHuang0302@163.com
EDUCATION 2017-2022
Architecture, Bachelor of Architecture in Architecture Shenzhen University, College of Architecture and Urban Planning
INTERNSHIPS 2021-2022
XKoolTech,Software Engineer, Research on Deep Learning in Architecture
2019
11 Architect,Research on Educational Architecture
EXPERIENCE 2021
2021 DigitalFUITURES: Workshops Artificial Intelligence & Big data Features prediction system Tongji Univerisity, Philip F.Yuan, Hao Zheng
2021
Kaggle – Shopee Price Matching Competition Data exploratory analysis and Multimodal prediction
20
2020 DigitalFUITURES: Workshops Convolution NeuralNetwork in Architectural Design Tongji Univerisity, Philip F.Yuan, Hao Zheng
HONORS 2021/2020
DigitalFUITURES: Workshops excellent team member
2021
Kaggle – Shopee Price Matching Competition Top11%
2020
Second Prize in National Scholarship
2019/2020
First Prize Excellent Student Cadre Scholarship
SKILLS Skilled in Python development and Deep learning applications Skilled in Computational design and Architectural robotics
LANGUAGE Native in Mandarin Chinese, Native in Cantonese, Fluent in English
CONTENT
P4 - 9
Multi-Modal Price Matching [Efficient-Net, NF-Net,Bert]
P10 - 17
Urban Big data Features Prediction [CNN,GAN,Big data Features Prediction System]
P18 - 25
Complex Façade Clustering [Museum Design,Machine Learning,Clustering]
P26 - 31
Market Memory Storage [YOLO_v3,Object Detection,Architecture Design]
P32 - 39
Topology Optimization Skyscraper
[Finite Element Analysis ( FEA), Structural Optimization]
01 Multi-Modal Price Match Guarantee Efficient-Net, NF-Net,Bert Kaggle : Shopee-Price Match Guarantee Teammate : Zoey Xu zoeyxu@aliyun.com Duration : 2021.5 Competition : Team Work (Leader)
Do you scan online retailers in search of the best deals?In this competition, we applied our machine learning skills to build a model that predicts which items are the same products. Captained my teammates to conduct data exploratory analysis, feature processing, building baseline and models optimization iteration. Responsible for parameters adjustment of Finetune and TFIDF for BERT model, and NFNet training for CV, the similarity algorithm selection, models’ fusion and prediction results splicing, ranked top 11% among over 3000 teams.
ABSTRACT Kaggle : Shopee-Price Match Guarantee Description : Retail companies use a variety of methods to assure customers that their products are the cheapest. Among them is product matching, which allows a company to offer products at rates that are competitive to the same product sold by another retailer. In this competition, well applied our machine learning skills to build a model that predicts which items are the same products. The applications go far beyond Shopee or other retailers. We contributions to product matching could support more accurate product categorization and uncover marketplace spam. Evaluation : F1 score = 2 × (Precision × Recall) / (Precision + Recall)
DATASETS 34250 (Image + Titles + Phash) 34250 images in the training set roughly 70,000 images in the hidden test set
[train/test].csv the training set metadata. posting_id - the ID code for the posting. image - the image id/md5sum. image_phash - a perceptual hash of the image. title - the product description for the posting. label_group - ID code for all postings that map to the same product. Not provided for the test set.
[train/test]images the images associated with the postings.(34250)
... × 34250
OVERVIEW My Solution in the Competition Top 11%
IMAGE Efficientnet B3 + Efficientnet B5 + Nfnet l0
TEXT BERT + TFIDF
PRECDITION KNN & Cosine Similarity
KNN -k-nearest neighbors algorithm
COMPARISON Comparison to Top 1% Kernel in the Competition
CONCLUSION Limitations & Future works
Future Works : After comparing with the winning kernel of the competition. The most serious deficiency of our kernel is post-processing and vote in prediction. According to the prediction results of the optimal LB score model in the early stage, we found that there are a large number of unmatched samples in the statistical results, that is, the matching results only match themselves, so we can reduce the matching standard (increase the matching distance threshold or reduce the similarity threshold) Comparison: Baseline ( TFIDF + ResNet18 + Result Splicing) : LB Score: 0.653 Top 80% My Solution (EfficientNet-B3 + EfficientNet-B5 + Nfnet-l0 + TFIDF + BERT + (Multi-Modal)) : LB Score: 0.734 Top 11% Top Solution (Nfnet-l0 + swin-small + EfficientNet-B0 + BERT + Indonesian-BERT + Multlingual-v1 + TFIDF) : LB Score:0.746 Top 1%
02 Urban Big data Features Prediction CNN,GAN,Big data Features Prediction System Projects in DigitalFUTURES and XkoolTech Tutor : Hao Zheng,Philip.F.Yuan,Tai Li Duration : 2020-2021 Design : 2.2/2.3 Individual Work 2.1 Team Work (Leader)
Description : Using the API of Amap to collect POI and geographic data of Shanghai, combined with the popularity and housing prices crawled of the Lianjia agent as basic dataset, adopted pix2pixHD model to establish forecast model for second-hand housing distribution map. Data: During this period, technologies such as data set visualization, feature engineering and data mining will be used, and finally the deep learning model will be input in the form of images.
ABSTRACT Urban Big data Features Prediction System
Description : Using the API of Amap to collect POI and geographic data of Shanghai, combined with the popularity and housing prices crawled of the Lianjia agent as basic dataset, adopted pix2pixHD model to establish forecast model for Popularity heatmap.
DATASETS POI Data + Graph Data
POI (Points of Interest) Data (From Amap.api)
Index : Index of second-hand houses in Shanghai.(Total 6000) location: Longitude and latitude of each sample. Hospnums: Number of hospitals within 1000m. Subnumbs: Number of Subway Stations within 800m.
Edubumbs: Number of Schools within 500m. Total_Price: The price of the whole house. Unit_Price: Unit price per square meter. Follow_Count: Number of followers on lianjia.com.
Househeat: Househeat(θ) = α / log2(β) α = Number of followers on lianjia.com. β = Duration on sale. *Numbs are Based on the service radius of each POI point. Graphs Data (From Baidu Map &ArcGis)
Architectural outline of Shanghai
Urban road Network of Shanghai
District Centroid Distance: Sum of coefficients of sample distance from each urban center. ( Closest 5 Urban Center)
OVERVIEW Data Processing + Generative Adversarial Networks (GAN)
ALGORITHM Generative Adversarial Networks (GAN)
Encoder
Decoder
Generator
Discriminator
DATA PROCESSING Data Processing of POI Data of Shanghai Central Area
Combie/Househeat
Combie/Unitprice
Hospital_numbs
Edu_numbs
District Centroid Distance
EDA and Structure Data
MODEL TRAINING Ⅰ Several different attempts of model training
image segmentation (Training set) Input Feature
Input Label
POI_Combination_Img
POI_Segmentation
Househeat_Img
Househeat_Segmentation
8000×8000
225×512×512
8000×8000
225×512×512
Model 01 input
POI_Image
real
Househeat_Image
synthesized
input
real
synthesized
epoch195 - epoch200
POI_Image: Subway only, 225 Segmentation
Result: Training set over fitting, Test set under fitting
Househeat_Image: Visualization based on thermal radius
Reason: Subway POI only is not comprehensive enough, and the visualization method of househeat has errors
Model: Pix2Pix, 200 epochs, Tesla V100
Model 02 input
POI_Image
Househeat_Image
real
synthesized
input
real
synthesized
epoch75 - epoch80
POI_Image: The urban map of green space and river is added
Result: Training set under fitting
Househeat_Image: Visualization based on thermal radius
Reason: The visualization method of househeat has errors,It's hard to connect with the urban map.
Model: Pix2Pix, 200 epochs, Tesla V100
MODEL TRAINING Ⅱ Several different attempts of model training
Model 03 input
POI_Image
real
Househeat_Image
synthesized
input
real
synthesized
epoch190 - epoch200
POI_Image: Full features, 50 selected Segmentation
Result: Training set over fitting, Test set under fitting
Househeat_Image: Visualization based on heat
Reason: The selected 50 fragments are easy to cause over fitting and poor generalization
Model: Pix2Pix, 200 epochs, Tesla V100
Model 04 input
POI_Image
real
Househeat_Image
synthesized
input
real
epoch09- epoch14
POI_Image: Full features, 225 Segmentation
Result: Training set under fitting
Househeat_Image: Visualization based on small radius
Reason: Too many black areas in the labels Delivered inaccurate information.
Model: Pix2Pix, 200 epochs, Tesla V100
synthesized
Model 05 input
POI_Image
Househeat_Image
real
synthesized
input
real
synthesized
epoch190 - epoch200
POI_Image: RGB_Full features, 225 Segmentation, with geo Mask.
Result: Good performance in training set and test set
Househeat_Image: Visualization based on Weighted value.
Reason: RGB color distinguishes different features well, which can also be learned by machines. Mask reduces interference in non residential areas
Model: Pix2Pix, 200 epochs, Tesla V100
PRECDITION Using model 5 to predict Househeat in Beijing
03 Complex Façade Clustering Museum Design,Machine Learning,Clustering
Longmenshan Ecological Diversity Museum Team Member: Xiaodi.Yang, Tai.Li, Kan.Liu, Changren.Wang, Rongxin.Liu Duration : 2021.3-2021.10 Xkool Technology Location : Sichuan China
Description : Longmenshan Ecological Diversity Museum is the result of an architectural competition I participated in during my internship xkool. It won the first place among 400 entries and will be built in Chengdu, Sichuan in 2022.I am mainly responsible for architectural generative algorithm design, architectural design, and standardization and clustering of Façade using machine learning. Generative Algorithm: We find the optimal building shape coefficient based on wind environment simulation, through millions of iterations. Façade Clustering: The construction of complex skin of buildings often needs to provide different molds for each panel. We hope to find the optimal panel clustering through K-means clustering, so that the same mold can correspond to different skins, taking into account both cost and architectural beauty.
ABSTRACT Longmenshan Ecological Diversity Museum Generative Algorithm: Longmenshan Ecological Diversity Museum is the result of an architectural competition I participated in during my internship xkool. We find the optimal building shape coefficient based on wind environment simulation, through millions of iterations.The workflow can also be applied to other scheme design processes that need to consider pedestrian comfort. Façade Clustering: The construction of complex skin of buildings often needs to provide different molds for each panel. We hope to find the optimal panel clustering through K-means clustering, so that the same mold can correspond to different skins, taking into account both cost and architectural beauty.
OVERVIEW Architecture Design + Façade Standardization Architecture Design
Façade Standardization
Input A Architectural Locations
Architecture Surface
UV Division
Initial Shape Coefficient
Output Surfaces
Panel Standardization
Programming in Python
K-means Clusting
RhinoCommon
Wind Simulation
No
Pedestrian Wind Comfort = Good
Calculate Costs = α
α < Thershold
Yes Yes
Output Modes and Panels
No
DESIGN BY NATURE Design based on wind environment simulation
Annual Wind Vectors
Pedestrian Wind Comfort
Simulation optimization curve
Results of ten iterations
Final optimization results
Pedestrian path
Museum Façade Design Nature Light + Nature Ventilation + Solar Power
Façade Detail Design Nature Light + Nature Ventilation + Solar Power
Façade division
Double Façade
Nature Ventilation
Façade Color
FACADE STANDARDIZATION the Same mold can correspond to different skins, Reduce Cost
1-Generate triangular mesh
2-Mesh Standardization
3-Geometry Features
6-Genetic algorithm iterative optimization
5-Calculate Costs
4-K-means Clusting
Non-Standardization
After-Standardization
Steps of Clusting (Panels in Thersholds)
Non-Clusting
After-Clusting
After-Clusting-Modes
Max-Divergence: 150mm Modes:25
Max-Divergence: 100mm Modes:55
Max-Divergence: 50mm Modes:128
Max-Divergence: 150mm Modes:25
Max-Divergence: 100mm Modes:55
Max-Divergence: 50mm Modes:128
LOST FUNCTION Future Works-Reference [Eigensatz et al. 2010] and review their main algorithmic
Divergence: quantifies the spatial gap between adjacent panels
kink angle: measures the jump in normal vectors between adjacent panels.
Surface fitting.: The deviation of the curve network, from surface F. Total: .
Curve fairness: Severity of tangential drift of the planels.
Panel centering: the mold center away from the segment center .
Max-Divergence: 10mm Max-KinAngle: 6°
04 Market Memory Storage YOLO_v3,Object Detection,Architecture Design
AI + Architecture Exhibition in Designsociety Tutor : Fei Qu fqu@szu.edu.cn Duration : 2021.12 Location : Shenzhen China Competition : Team Work (Leader)
Design society, china’s first dedicated cultural design hub, has opened its doors to the public in shenzhen. the new institution is housed within the sea world culture and arts center (SWCAC), a building designed by japanese architect fumihiko maki, and houses a new V&A gallery. Shekou fishing Market will be demolished by the end of November 2020. Invited by Design society and Huawei cloud , we will build an art installation in the sea world culture and arts center (SWCAC) to continue Shekou people's memory of Shekou fishing Market. This art installation combining AI and architecture will be displayed in the museum. We will place the camera in the market and use Yolo_v3 algorithm carries out human shape detection in real time. Through stepping motors and art devices in the museum, we conveys the same sense of crowding to exhibitors as in Shekou market. This sense of crowding will also be preserved in the form of data.
ABSTRACT AI × Architecture Interactive art installation Description : Shekou fishing Market will be demolished by the end of November 2020. Invited by Design society and Huawei cloud , we will build an art installation in the sea world culture and arts center (SWCAC) to continue Shekou people's memory of Shekou fishing Market. This art installation combining AI and architecture will be displayed in the museum. Interactive: We will place the camera in the market and use Yolo_v3 algorithm carries out human shape detection in real time. Through stepping motors and art devices in the museum, we conveys the same sense of crowding to exhibitors as in Shekou market. This sense of crowding will also be preserved in the form of data.
SHEKOU Design Society Museum © Maki and Associates
Shekou Market since 1956
Design Society Museum since 2017
CAMERA LOCATION Three Cameras in Shekou Market to record People's activities
OVERVIEW Market + Algorithm + Museum
ALGORITHM YOLO_v3 Object Detection
DBL×5
Darknet-53 without FC Layer DBL
Res1
Res2
Res8
Res8
Res4
DBL
DBL
416×416×3
DBL
conv
Up Sampling
concat
DBL
DBL
conv
DBL×5
Yolo_v3_Structure
DBL
Up Sampling
concat
DBL
=
conv
BN
Res_unit Leaky relu
res unit
Y2 26×26×255
DBL
DBL
conv
DBL×5
Darknetconv2D_BN_Leaky
Y1 13×13×255
Y3 52×52×255
Resblock_body
=
DBL
DBL
Add
resn
=
Zero Padding
DBL
res unit Res_unit×n
YOLO_v3 Algorithm Structure
Video of Shekou Market
Yolo_v3 Loss Function ©Keyird
STRUCTURE Elastic Fabric + Stepper Motor + Prefabricated Frame
EXHIBITION Exhibition from December 2020 to February 2021
MOTION Controled by Programmable Logic Controller (PLC)
PLC Control the movement of the Stepper Motor
Market - Museum Movement in each time period
2D Plan Drawing of PLC
05 Topology Optimization Skyscraper Finite Element Analysis ( FEA), Structural Optimization
Tutor : Wenhua Deng , Mike Xie Duration : 2020.9 - 2020.12 Competition : Individual Work
Description: In this Skyscraper design, I will challenge the height of 600m. Higher buildings mean higher structural strength. Inspired by the One Thousand Museum designed by Zaha in Miami, I used the beso algorithm studied by RMIT Dr. Mike Xie to optimize the topology of building structures. BESO: BESO can add elements in the area of material "high efficiency" (such as high material stress). This ability to add elements makes the shape optimization ability of progressive optimization method stronger, easier to reduce the maximum stress and stress concentration, make the structural stress distribution uniform, and find a better force transmission path. It can also delete inefficient units in the model. We use the stiffness optimization mathematical model to illustrate the basic principle of beso. Firstly, taking the element itself as the design variable and the average compliance of the structure as the objective function, we establish the mathematical optimization model under volume constraints.
ABSTRACT Finite Element Analysis ( FEA), Structural Optimization
Description: In this Skyscraper design, I will challenge the height of 600m. Higher buildings mean higher structural strength. Inspired by the One Thousand Museum designed by Zaha in Miami, I used the beso algorithm studied by RMIT Dr. Mike Xie to optimize the topology of building structures. BESO: BESO can add elements in the area of material "high efficiency" (such as high material stress). This ability to add elements makes the shape optimization ability of progressive optimization method stronger.. It can also delete inefficient units in the model. We use the stiffness optimization mathematical model to illustrate the basic principle of beso. Firstly, taking the element itself as the design variable and the average compliance of the structure as the objective function, we establish the mathematical optimization model under volume constraints.
OVERVIEW Environmental Simulation + Structure Computing + Architecture Design
Environmental Simulation
Structure Computing
Optical Simulation Thermal Simulation
External horizontal load
Wind Simulation Visual Simulation
Architecture Design
Site Design Change Flow Condition
CFD
BESO
Give Load Parameters
Frame-tube Design
Internal vertical load
Functional Design Interior design
SITE DESIGN Efficientnet B3 + Efficientnet B5 + Nfnet l0
Mechanical Simulation Site Design The traditional urban design is a uniform grid, and the shortest path is monotonous. We hope to get the best path through the analysis of human flow mechanics.
8:00 am
10:00 am
2:00 pm
4:00 pm
Ground Floor Plan
B1 Floor Plan
B2 Floor Plan
B3 Floor Plan
6:00 pm
Structure Evolution
Bi-directional Evolutionary Structural Optimization (BESO)
10KN
Symmetry: Yes Volume Fraction: 0.5 Total Energy(MPa): 0.612 Total Epochs: 138
Epochs 5
Epochs 65
Epochs 138
Epochs 5
Epochs 95
Epochs 191
Epochs 5
Epochs 50
Epochs 106
Epochs 5
Epochs 45
Epochs 93
Epochs 5
Epochs 40
Epochs 74
10KN
Symmetry: No Volume Fraction: 0.2 Total Energy(MPa): 0.021 Total Epochs: 191
10KN
Symmetry: No Volume Fraction: 0.2 Total Energy(MPa): 0.007 Total Epochs: 106
10KN Symmetry: No Volume Fraction: 0.5 Total Energy(MPa): 150.547 Total Epochs: 93
10KN Symmetry: No Volume Fraction: 0.5 Total Energy(MPa): 57.663 Total Epochs: 74
Structure Evolution
Bi-directional Evolutionary Structural Optimization (BESO)
10KN
10KN Symmetry: No Volume Fraction: 0.5 Total Energy(MPa): 10.746 Total Epochs: 300
Epochs 5
Epochs 150
Epochs 300
Epochs 5
Epochs 43
Epochs 86
Epochs 5
Epochs 76
Epochs 157
Epochs 5
Epochs 32
Epochs 64
Epochs 5
Epochs 37
Epochs 74
10KN
Symmetry: No Volume Fraction: 0.4 Total Energy(MPa): 28.671 Total Epochs: 86
5KN
10KN
10KN
Symmetry: No Volume Fraction: 0.4 Total Energy(MPa): 0.218 Total Epochs: 157
10KN
10KN
Symmetry: No Volume Fraction: 0.4 Total Energy(MPa): 0.053 Total Epochs: 64
10KN Symmetry: No Volume Fraction: 0.5 Total Energy(MPa): 58.635 Total Epochs: 74
10KN
BESO Bi-directional Evolutionary Structural Optimization Method Description : BESO can add elements in the area of material "high efficiency" (such as high material stress). This ability to add elements makes the shape optimization ability of progressive optimization method stronger, easier to reduce the maximum stress and stress concentration, make the structural stress distribution uniform, and find a better force transmission path. It can also delete inefficient units in the model. We use the stiffness optimization mathematical model to illustrate the basic principle of beso. Firstly, taking the element itself as the design variable and the average compliance of the structure as the objective function, we establish the mathematical optimization model under volume constraints. V*=∑Vixi, xi=0 or xi=1
F and U are the structural load matrix and displacement matrix respectively, C is the average strain energy of the structure, V * and VI are the target volume and element volume in turn, and Xi is the design variable.
Sensitivity Filtering
Typical convergence process
PHASE I Optimize the structure using the first generation model
Iterative Process
Non-Design
Horizontal Load
Vertical Load
Support
Phase I
PHASE Ⅱ Integration Model after Subsection Optimization
Split Brep
Mesh
Support
H.Load
V.Load
Mirror
Typical convergence process
Phase II
PHASE Ⅲ Structural optimization of first floor Theater and Shopping Mall
Theater (Axon)
Mall-1 (Axon)
Mall-2 (Axon)
Theater (Section)
Mall-1 (Section)
Mall-2 (Section)
Top view
East view
Phase III