/wandb-mlops

A project of Semantic Segmentations Modelling over Cityscape Datasets that incorporating Workflows, Experiment Tracking, Pipeline and testing

Primary LanguagePython

BDD Semantic Segmentation using WandB

This is semantic segmentation project on BDD cityscapes-like dataset that are tracked via wandb. You can see the run on my wandb.

The objectives of this project is to segment pedestrian view of car camera view for semantic segmentation problem using wandb to track and run the code modularly with this repository to ensure the project reproducible.

How to set up the project

The setups are pretty basic, just go

> pip install virtualenv
> virtualenv venv
> source venv/bin/activate
> pip install -r requirements.txt
> wandb login

Then run the code sequentially

> python data_loader.py
> python split.py
> python baseline.py
> python eval.py

You only need to run data_loader.py and split.py once since this data is static, but you could run baseline.py and eval.py multiple times since it is what the experiments about.

You can set configs of hyperparameters to experiment in the baseline.py file, try tweak some of the hyperparameters. The eval.py is to check the model that are produced on baseline.py executions on test holdout set.

Reports on Wandb

You can check my reports on this repository executions on these pages