(Day 37) PyTorch traffic sign classification and detection model

Ivan Ivanov · February 7, 2024

Hello :) Today is Day 37!

A quick summary of today:

Firstly

to put knowledge into practice I decided to try to make an animal classification model

There are some limitations with Kaggle’s GPU so I had to tone down the picture sizes, the augmentations and model complexity.

So ~ data augmentation

image

Image example:

image

The categories: cat, dog, elephant, horse, lion

I divided the data into batches:

image

The model is:

image image

The training took a lot of time; below you can see the last epoch’s results

image

Example classification:

image image image

(there are some weird classifications too haha)

**Next ~ **

About the traffic sign classification and localisation model

Here is a link to the notebook on Kaggle.

The model’s dataset has 43 classes (here are 21 of them)

image image

The image size varied a lot so I standardised it

image

Model architecture:

image image image

I tried various architectures, but the accuracy was not that good, so online I found AlexNet and wow ~

image

Example:

image

I applied what I learned yesterday to this model according to the vegetable/apple bounding box model from yesterday, but the accuracy and loss won’t decrease even after training, and I’ll try again tomorrow!!!

Now it’s 2.37am and I will go to bed ~

That is all for today!

See you tomorrow :)