This is a simplt but useful Pytorch CNN binary classifier walk through. See the detail of content in the .pptx (in chinese).
If anyone needs English version of the slide, please let me know <3
- How to use simple chrome extension, to web crawl images down.
- How to write
datasetclass with appropriate labeling and sorting, and data augmentation. - How to use a pretrained model from Pytorch official package, and also how to build a model for yourself. Furthmore, I write a simple counter for computing number of neurons for the fully connect layer input.
- How to write a reader friendly log for future examination.
- How to use tensorboard in Pytorch
- Comprehensive slides explain every step of building your own project.
- Detail comments along the side of code.
- Introduce some useful packages, for better training inspectation.
- Some tips and tricks for both preprocess and postprocess
In the end of All_in_one.ipynb, I provide visualization of model results (Grad-CAM) and metric (confusion matrix).