Studijní materiály (lectures)
1 - Intro: Autonomous Vehicle, Object Detection, Image Features1.1 face-haar-example.zip
1.2 sliding_window.mp4
1.3 AdaBoost.xlsx
2 - Convolutional networks: LeNet, PyTorch
2.1 nn-from-scratch-pytorch.py
2.2 nn-simple-classification-model.py
2.3 sgd-example.py
2.4 loss-example.py
3 - CNN Architectures: AlexNet, ZFNet, VGG, GoogLeNet, ResNet, DenseNet
3.1 Dropout, Pytorch Dataset
3.2 ResNet, SqueezeNet, EfficientNet
3.3 feature-map-visu.py
3.4 dropout-example.py
3.5 batch-norm_example.py
3.6 1x1-example.py
3.7 3x3x3-1x7x7-example.py
4 - Object Detection: RCNN, YOLO, SSD
4.1 car_detection.mp4
4.2 RCNN.zip
5 - ViT (Vision Transformer)
6 - Generative Adversarial Networks (GANs)
6.1 parking_gan.mp4
6.2 parking_dcgan.mp4
Cvičení (exercises)
| Name | Description | References |
|---|---|---|
| 1. Parking Lot Analysis (6p) | Experiment with classical edge detection, Haar/HOG/LBP for parking lot analysis. |
slides video template.zip scikit-image trainimages |
| 2. Neural Network - Pytorch (6p) | Experiment with neural networks for parking lot analysis (use data from from the previous exercise). You can combine the following files to create a basic classifier. |
01-torch-dataset.py 02-torch-model0.py 03-torch-trainloop.py |
| 3. CNN - Pytorch (12p) | CNN for parking lot analysis (use data from from the previous exercise). Experiment with different CNN architechtures (e.g. VGG, ResNet, GoogLeNet, ViT - vision transformer). Try to design (implement) and test your own architecture (e.g. My-VGG, My-ResNet, My-GoogLeNet, My-ViT). Experiment with layers, learning rate, number of epochs, normalization, dropout, loss function, data augmentation. Create a report (e.g. document with graphs, tables, configurations) that maps the performed experiments (max. 5 pages). |
cifar10_tutorial alexnet.py vgg.py inception.py resnet.py tensorboard.py netron.py t-learning.py |
| 4. RCNN/SSD/RetinaNet/YOLO (6p) | Experiment with different detection approaches to improve parking space analysis. You can use already trained (available) models - but try to combine the detection results with the recognition results (from the previous exercises) to produce the output prediction. |
Pytorch Object Detection Ultralytics YOLO11 |
| 4.1 Custom YOLO Training (5p) | + The addition to the previous exercise 4: Experiment with different yolo parameters to train your own YOLO model for vehicle detection to improve parking space analysis. For annotation, you can use the annotation program included in yolov11.zip archive or some other ways (e.g. https://www.makesense.ai/). For the training data, you can use attached train-images-for-yolo.zip |
yolov11.zip train-images-for-yolo.zip example-of-train-batch0.jpg |
| 6. GAN (5p) | Experiment with GAN to extend the parking lot training dataset. Compare results with and without the extended set. |
video DCGAN Tutorial |
| 10.12.2024 12.12.2024 |
personal presentations of the tasks |
