| 週次 |
授課內容 |
| 第1週 |
Course overview, learning outcomes, image representation, pixels, RGB/gray scale, histograms, convolution, common vision tasks, Python/Colab/OpenCV warm-up. |
| 第2週 |
AI/ML/DL relationship, data-objective-model-metric-optimization pipeline, supervised learning, linear and nonlinear regression, loss functions, learning rate, gradient descent. |
| 第3週 |
Underfitting/overfitting, entropy, information gain, regression trees, pruning, clustering vs. classification, density estimation, hierarchical clustering, k-means, PCA visualization. |
| 第4週 |
Numeric representation, tensors, neural network architecture, hidden layers, activation functions, learning modes, GPU/TPU concepts, MNIST introduction. |
| 第5週 |
Artificial neurons, bias, sigmoid, ReLU, perceptron, hidden layers, PyTorch/Keras MNIST implementation, epochs, layers, activation functions, dropout. |
| 第6週 |
DNN depth, parameter explosion in images, CNN locality, weight sharing, convolution, stride, padding, pooling, flattening, LeNet, AlexNet, VGG, CIFAR-10. |
| 第7週 |
Sequence data, hidden state, one-hot encoding, RNN limitations, vanishing gradient, LSTM gates, lookback windows, wind speed and traffic prediction examples. |
| 第8週 |
Midterm Exam. |
| 第9週 |
Bounding boxes, class labels, confidence scores, IoU, mAP, annotation quality, YOLO training workflow, dataset format. |
| 第10週 |
Semantic segmentation, instance segmentation, masks, Dice score, mIoU, U-Net concept, medical/industrial/remote-sensing examples. |
| 第11週 |
Pretrained models, feature extraction, fine-tuning, augmentation strategies, train/validation/test leakage, class imbalance. |
| 第12週 |
Confusion matrix, precision, recall, F1, ROC/AUC, mAP review, overfitting diagnosis, failure-case visualization, reproducibility. |
| 第13週 |
Frame-based processing, motion cues, object tracking, ID switches, MOTA/IDF1/HOTA concepts, traffic and surveillance applications. |
| 第14週 |
Vision Transformers, multimodal vision-language models, image restoration, anomaly detection, generative AI applications, ethical and data-source issues. |
| 第15週 |
Student presentations on selected image processing or deep learning applications, including problem definition, data, model, evaluation, and limitations. |
| 第16週 |
Final Exam. |
自主學習
內容 |
02.閱覽產業及學術相關多媒體資料
|
|