Cannabis Data Science

3D Body Scan Clothing Size Classification

Contracted to an early stage startup focusing on clothing size recommendations using 3D body scans.

Researched the use of 1D convolutional neural networks and transformers, to classify the clothing size of an individual using point clouds obtained from LiDAR scans.

Multimodal neural network research involving convolutional and multilayer neural networks to combine 2D image and tabular data to make clothing size classification predictions.

Gradient Boosted Decision Tree women’s clothing size classification research using measurements derived from 3D LiDAR body scans. Achieved 80%+ accuracy on sizes where sufficient data existed. XGBoost, CatBoost, LightGBM, Python, Pandas, Optuna, Scikit-Learn

Engineered new features that improved prediction accuracy including waist to height ratio that accurately measures body fat distribution.

Performed exploratory data analysis, outlier removal, highlighted data deficiencies.

Improved the accuracy of a XGBoost men’s clothing size classifier from 85% to 99% that utilized measurements from 3D LiDAR body scans.

Technology

  • PointNet
  • Point Cloud Transformer
  • Python
  • Pandas
  • Keras
  • TensorFlow
  • TriMesh
  • Open3D
  • PyTorch
  • OpenCV
  • EfficientNet
  • Transfer Learning
  • Jupyter Notebook
  • Scikit-Learn
  • NumPy
  • XGBoost
  • Catboost
  • LightGBM
  • Optuna
  • Matplotlib
  • Seaborn
  • Pandas Profiling