Tutorial on Image Classification using Scikit-Image, Scikit-learn, and PyTorch
summary
Pipeline that actually runs: scikit-image for segmentation (adaptive histogram, multi-Otsu, remove_small_objects, morphological ops and a WA segmentation workflow), feature extraction with GLCM and LBP into ~16 features, and a Gaussian process classifier trained on ~200 images per class that reports strong precision/recall; PyTorch CNNs (simple CNN and ResNet-50 transfer learning) with augmentation, TensorBoard logging, saliency maps, and notes about needing a GPU for ResNet training. Code and dataset links live in the GitHub repo.
AI-written summary · about
chapters (21)
- 0:00Introduction and speaker overview
- 1:05Agenda and image classification pipeline
- 3:15Overview of image processing tools
- 4:40Image segmentation fundamentals
- 8:53Thresholding and morphological operations
- 11:03Live demo of image segmentation
- 19:13Advanced segmentation with Gaussian mixture models
- 21:25WA segmentation with random forest classifier
- 24:14Feature extraction for virus classification
- 27:26Gaussian process classification explained
- 29:17Demo of Gaussian process classification
- 38:31Key performance metrics for classification
- 38:51Introduction to PyTorch and CNNs
- 40:37Building a simple CNN architecture
- 43:27Training loop and TensorBoard setup
- 46:25Data augmentation techniques
- 51:20Model initialization and training process
- 53:35Monitoring metrics with TensorBoard
- 55:50Salency maps for CNN interpretation
- 59:38Transfer learning with ResNet
- 1:03:36Summary and Q&A session
related talks
45mTech Tutorial: Image segmentation
01h 27mHands-On Computer Vision with PyTorch
59mBuild and deploy production ready PyTorch models
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
01h 07mComputer Vision with PyTorch
51mA Practical Introduction to Convnets for Image Classification
01h 32mIntroduction to PyTorch
01h 31mPyTorch Workflow Mastery: A Guide to Track and Optimize Model Performance