About Computer Vision
Computer vision is the field of teaching machines to understand images, video, and
3D scenes, and this course walks through it in the order most practitioners
actually learn it. It starts with Image Fundamentals (pixels, color
spaces, and OpenCV), then Classical Computer Vision and
Deep Learning Foundations so you understand both the
hand-crafted methods and the neural networks that replaced them.
Next comes Image Classification & Backbones: CNNs, Vision
Transformers, and transfer learning. From there,
Object Detection covers YOLO, Faster R-CNN, and DETR, while
Segmentation covers U-Net, Mask R-CNN, and promptable models like
the Segment Anything family.
Vision-Language & Generative Models covers CLIP-style
embeddings, multimodal models, and diffusion, and
Video Understanding covers optical flow, multi-object tracking,
action recognition, and video transformers.
Finally, 3D Vision covers depth estimation, point clouds,
structure from motion, NeRFs, and Gaussian splatting, and
Evaluation & Deployment covers metrics, optimization, and
serving models in production: the skills that separate a research demo from a
vision system that works on real-world data.
Content
10 modules
01
Image Fundamentals
How images are represented and manipulated: pixels, channels, color spaces, resizing, and reading and writing images with Python.
NumPyOpenCVColor Spaces
02
Classical Computer Vision
The pre-deep-learning toolkit: filtering, edge and corner detection, local features, and the geometry of cameras.
ConvolutionSIFT / ORBCamera Geometry
03
Deep Learning Foundations
Neural networks, backpropagation, and training loops in PyTorch, plus data pipelines and augmentation for vision tasks.
PyTorchCNNsData Augmentation
04
Image Classification & Backbones
From ResNet to Vision Transformers: how backbone architectures work and how to fine-tune pretrained models on your own data.
ResNetVision TransformerTransfer Learning
05
Object Detection
Locating and labelling multiple objects in an image with one-stage, two-stage, and transformer-based detectors.
YOLOFaster R-CNNDETR
06
Segmentation
Pixel-level understanding: semantic, instance, and panoptic segmentation, including promptable foundation models.
U-NetMask R-CNNSegment Anything
07
Vision-Language & Generative Models
Connecting images and text with joint embeddings and multimodal models, and generating images with diffusion.
CLIPMultimodal LLMsDiffusion
08
Video Understanding
Modelling motion and time: optical flow, object tracking across frames, action recognition, and video transformers.
Optical FlowObject TrackingAction Recognition
09
3D Vision
Recovering depth and structure from images: monocular depth, point clouds, structure from motion, and neural scene representations.
Depth EstimationPoint CloudsNeRF / Gaussian Splatting
10
Evaluation & Deployment
Measuring model quality with the right metrics, then optimizing and serving vision models on servers, edge devices, and in the browser.
mAP / IoUONNXTensorRT
What You Can Do After This Course
By the end of the curriculum you'll have worked through 5 hands-on projects and
covered the core of what a junior to mid-level computer vision role expects.
Concretely, you'll be able to:
Work with visual data: load, transform, and augment images and video, and explain how a camera turns a scene into pixels.
Train and fine-tune models: build vision networks in PyTorch and adapt pretrained CNN and transformer backbones to your own dataset.
Detect and segment: find objects with YOLO or DETR-style detectors and produce pixel-accurate masks with U-Net, Mask R-CNN, or promptable models.
Combine vision and language: use joint embeddings and multimodal models for search, captioning, and visual question answering, and understand how diffusion models generate images.
Understand video: track objects across frames, estimate motion, and recognize actions with temporal models.
Reason in 3D: estimate depth, work with point clouds, and reconstruct scenes using structure from motion, NeRFs, and Gaussian splatting.
Ship it: evaluate with the right metrics, optimize models with ONNX or TensorRT, and deploy them reliably.