Applied Computer Vision Essentials Training in Switzerland

Who Should Attend?

• Working knowledge of Python 3.9+: functions, classes, virtual-environment management (venv or conda), package install with pip.
• Familiarity with NumPy arrays and tensor concepts; ability to write a simple forward pass in PyTorch or TensorFlow.
• Experience running a supervised-learning loop: dataset split, loss calculation, back-prop, checkpoint save.
• Basic shell skills on Linux (navigate directories, edit config files, run git clone).
• Git fundamentals: clone, branch, commit, push, pull-request workflow.
• JupyterLab usage: open notebooks, run cells, inspect GPU memory.
• Awareness of GPU vs CPU execution; can read nvidia-smi output or fallback to CPU when GPUs are unavailable.
• Introductory linear-algebra and probability: matrix multiply, softmax, cross-entropy.
• Ability to read JSON/YAML config files and tweak hyper-parameters.
• Laptop or desktop with stable broadband (≥ 10 Mbps down / 2 Mbps up) and a modern browser that reaches Skillable lab URLs over HTTPS.
• Company VPN, proxy, or security policy allows outbound WebSocket traffic for JupyterLab (ports 8888/8443) and VS Code Server if used.
• Optional but helpful: basic Docker commands (docker build, docker run) and REST API testing with curl or Postman.

• Apply classical computer vision techniques for edge detection, feature extraction, and lane detection
• Analyze color spaces, histogram equalization, and contrast enhancement methods for image quality improvement
• Create data augmentation pipelines and fine-tune CNN architectures like EfficientNet for classification
• Evaluate object detection performance using mAP and IoU metrics with TIDE error analysis
• Implement YOLO training workflows for safety compliance with hyperparameter optimization
• Compare segmentation approaches from traditional methods to modern promptable SAM 2
• Construct Vision Transformer solutions using DINOv2 and self-supervised learning principles
• Synthesize multimodal pipelines integrating detection, CLIP embeddings, and language models for alt-text generation
• Optimize models for production through ONNX conversion, INT8 quantization, and edge deployment
• Assess computer vision systems for bias and fairness while implementing production monitoring with Prometheus

• Pixels, color spaces, convolution filters
• Lane‑finding with Canny + Hough
• Histogram equalisation & CLAHE
• Low‑light rescue with CLAHE
• Feature extraction: classical descriptors
• Image matching: ORB vs SIFT
• CVAT annotation + COCO export
• Wrap-up: bridging classical to modern CV

• Classical to deep transition
• CNN architectures & evolution
• Data‑augmentation strategies
• AutoAugment & RandAugment demo
• Fine‑tune EfficientNet‑V2‑S + Grad‑CAM
• Intro to object detection & YOLO family
• YOLOv11‑nano training start
• Detection metrics & interpretation; TIDE taxonomy
• Model robustness discussion

• From detection to segmentation
• Segmentation approaches
• SAM 2: promptable segmentation
• SAM 2 segmentation vs YOLO masks
• Vision Transformers revolution
• Video processing fundamentals
• Attention rollout visualisation
• Self-supervised learning
• Fine‑tune DINOv2‑tiny
• Modern CV landscape
• Capstone prep

• Recap: CV evolution journey
• Vision-language models
• Image & video generation
• Detector → CLIP → LLM safety report
• Model deployment essentials
• ONNX conversion & optimization
• Production monitoring demo
• Adversarial robustness
• Ethics in Computer Vision
• Wrap-up; Q&A
• Capstone demos