Syllabus

Logistic, Introduction, History

Image processing: image gradient, filter, convolution.

Classic edge/corner/line detection methods: Canny edge detection, Harris corner detection, line fitting.

Edge/line detection: Canny edge detection, line fitting(RANSAC, Hough transform).

Corner detection and feature descriptors: Harris corner detection, feature.

Outline of deep leanring.

Outline of deep leanring.

Network structure: Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN).

CNN Training: weight initialization, loss, optimizer (gradient descent, SGD, Adam), learning rate.

Underfitting and overfitting: batch normalization, skip link, augmentation, regularization

Classification task: KNN, SoftMax, cross entropy loss

Classification task: receptive field, architecture(VGG, ResNet…)

Segmentation task: K-Means, upsampling, architecture (FCN, UNet)

Camera model: pinhole camera, lense, intrinsic, extrinsic, calibration

Depth image: sensor, stereo

3D data (voxel, mesh); 3D Deep Learning: Point, Mesh; 3D Deep Learning: Sparse Voxel Conv

Motion and optical flow

RNN, LSTM, GRU; Video Analysis: 3D CNN

Attention mechanism, transformer, visual transformer, vision language pretraining, large language model, vision language model.

2D Object detector (SSD, RCNN series, YOLO); Instance Segmentation, Panoptic Segmentation; 3D object detection and instance segmentation

VAE, GAN, Diffusion, text-to-image generation.

Object grasping, object manipulation, locomotion and navigation, embodied vision-language models.