Programming Tips

Thoughts on CosineEmbeddingLoss

pytorch

\[ \begin{align} \text{loss}(x, y) = \begin{cases} 1 - \cos(x_1, x_2) & \text{if } y = 1 \\ \max(0, \cos(x_1, x_2) - \text{margin}) & \text{if } y =…

Pydantic Post Init

python

If you want to update an existing property, you can do this:

Timm choosing a model

pytorch

Choosing a backbone for transfer learning is easy thanks to Timm. However, which model is a mystery. I use the following to use a small model which has >80% accuracy on…

Pytorch Lightning Learning rate finder

pytorch

The following plots the loss against learning rate in order to discover an “optimal” learning rate.

An Intuitive Loss for Imbalanced Classification

An alternative approach to Focal Loss

Coco Semantic Segmentation in PyTorch - Data Prep

pytorch

data

How to prepare and transform image data for segmentation

Sachin Abeywardana