Studijní materiály (lectures)
1 - Intro: Autonomous Vehicle, Object Detection, Image Features
1.1 Appendix: sliding_window.mp4
2 - Convolutional networks: LeNet, PyTorch
3 - CNN Architectures: AlexNet, ZFNet, VGG, GoogLeNet, ResNet, DenseNet
3.1 - Appendix: Dropout, Pytorch Dataset
3.2 - Appendix: ResNet, SqueezeNet, EfficientNet
4 - Object Detection: RCNN, YOLO, SSD
4.1 Appendix: car_detection.mp4
5 - Generative Adversarial Networks (GANs)
5.1 Appendix: parking_gan.mp4
5.2 Appendix: parking_dcgan.mp4
Cvičení (exercises)
| Name | Description | References |
|---|---|---|
| 1. Parking Lot Analysis | Experiments with classical edge detection, or Haar/HOG/LBP for parking lot analysis. |
video template.zip scikit-image trainimages |
| 2. Car Detection - Sliding Window | Experiments with HOG and car localization. Create a program using a sliding window in combination with the HOG features. | video vehicle_data neg-bmw pos-bmw hog-scikit SVM sklearn |
| 3. PyTorch/NN | Link all parts together and add a testing phase and display the results using matplotlib. |
PyTorch_dataset PyTorch_model_0 PyTorch_train |
| 4. CNN - Pytorch | Experiments with CNN and parking lot analysis. For homework, use trainimages (for training) and big parking images from template.zip (for testing) | tutorial |
| 5. RCNN/YOLO/SSD | Experiments with RCNN (YOLO/SSD) - In the first step, the goal is to find all cars using RCNN methods (or YOLO/SSD). In the second step, use CNN for recognition of specific car (e.g. BMW). You can use training images from previous tasks or create new ones. vehicle_data |
video slides yolov8_example.zip |
| 6. GAN | Use GAN to extend the dataset in one of the above examples. Compare results with and without the extended set. | video |