The best way to internalize the knowledge on these pages is to implement it. What does this mean?
- For architectures: build them from scratch.
- For training paradigms: test out different methods and get an intuitive feel for the results.
- For theorems/generally accepted rules: derive them from scratch
With that in mind, here are implementations I’d recommend you take on to truly cement the learnings from this section on Computer Vision.
In the panoramic landscape of modern technology, computer vision is drawing inspiration from the biological intricacies of the human eye and brain, to endow machines with the ability to decipher images and videos, transcending mere pixel analysis to derive meaningful interpretation.
The content
Classifiers (numpy)
- K Nearest Neighbors
- Support Vector Machine Classsifier
- Softmax Classifer
- Multilayer Neural Network
CNNs + NNs (numpy)
- Multil-layer Fully Connected Networks
- Batchnorm
- Dropout
- CNN Layers
Intro to PyTorch (by stages)
- Barebones: low-level pytorch tensor
- PyTorch Module API:
nn.Moduleto define arbitrary nn architectures - PyTorch Sequential API:
nn.Sequential
Applications / Advanced Architectures