Computer Vision

Notes based on lectures by Dr. Yuqi Ouyang (SCUPI), CS1674 Introduction to Computer Vision.

Computer vision teaches machines to perceive and understand the world from images and video — from low-level image processing (filtering, resampling, features) through geometry-based vision (homography, stereo, motion, tracking) to modern deep learning (CNNs, RNNs, vision transformers, generative models) and high-level recognition.

Chapters

• Chapter 1: Introduction — What CV is, goals, applications, and why it is hard (variation, ambiguity, context)
• Chapter 2: Image Filtering — Images as functions, convolution/correlation, mean/median/Gaussian filters, noise
• Chapter 3: Image Resampling — Aliasing, Nyquist theorem, Gaussian prefiltering, pyramids, interpolation/upsampling
• Chapter 4: Feature Detection — Edges, Sobel, Canny, line/circle fitting, Hough transform, RANSAC, Harris corners
• Chapter 5: Feature Description and Matching — SIFT descriptors, invariance, matching, ratio test, visual words / bag-of-words
• Chapter 6: Homography and Projective Transformation — 2D transforms, homogeneous coords, homography, panorama stitching
• Chapter 7: Stereo Vision — Epipolar geometry, rectified stereo, disparity and depth
• Chapter 8: Motion Estimation — Optical flow, Lucas-Kanade, brightness constancy, coarse-to-fine
• Chapter 9: Tracking — KLT tracking, mean-shift, Kalman filtering
• Chapter 10: Introduction to Deep Learning — MLPs, activation functions, universality of ReLU networks
• Chapter 11–12: Training Deep Learning Models — Gradient descent, SGD, backprop, optimizers (momentum, RMSProp, Adam), regularization, LR decay, early stopping
• Chapter 13: Encoder-Decoder for Latent Analysis — Representation learning, autoencoders, bottleneck, sparse/contractive AEs
• Chapter 14: Convolutional Neural Networks — Convolutions, pooling, 1×1/depthwise, backprop, AlexNet/VGG/GoogLeNet/ResNet
• Chapter 15: Recurrent Neural Networks — Sequence models, vanilla RNN, BPTT, LSTM/GRU, image captioning
• Chapter 16: Vision Transformer — Attention, transformers, ViT, Swin, DINO, MAE, SAM
• Chapter 17: GANs — GAN objective, DCGAN, evaluation (IS/FID), pix2pix, CycleGAN
• Chapter 18: Diffusion Models — DDPM, score-based, classifier-free guidance, latent diffusion, DDIM, SDXL, LCM
• Chapter 19: Object Recognition — Detection/segmentation tasks, HOG, DPM, R-CNN family, YOLO, FCN, Mask R-CNN, CLIP