Lightning Kitti

Semantic Segmentation with Pytorch-Lightning

Introduction

This is a simple demo for performing semantic segmentation on the Kitti dataset using Pytorch-Lightning and optimizing the neural network by monitoring and comparing runs with Weights & Biases.

Pytorch-Ligthning includes a logger for W&B that can be called simply with:

from pytorch_lightning.loggers import WandbLogger
from pytorch_lightning import Trainer

wandb_logger = WandbLogger()
trainer = Trainer(logger=wandb_logger)

Refer to the documentation for more details.

Hyper-parameters can be defined manually and every run is automatically logged onto Weights & Biases for easier analysis/interpretation of results and how to optimize the architecture.

You can also run sweeps to optimize automatically hyper-parameters.

Note: this example has been adapted from Pytorch-Lightning examples.

Usage

Notebook

• A quick way to run the training scrip is to go to the notebook/tutorial.ipynb and play with it.

Script

1. Clone this repository.

2. Download Kitti dataset

3. The dataset will be downloaded in the form of a zip file namely data_semantics.zip. Unzip the dataset inside the lightning-kitti/data_semantic/ folder.

4. Install dependencies through requirements.txt, Pipfile or manually (Pytorch, Pytorch-Lightning & Wandb)

5. Log in or sign up for an account -> wandb login

6. Run python train.py and add any optional args

7. Visualize and compare your runs through generated link

   

Sweeps for hyper-parameter tuning

W&B Sweeps can be defined in multiple ways:

• with a YAML file - best for distributed sweeps and runs from command line
• with a Python object - best for notebooks

In this project we use a YAML file. You can refer to W&B documentation for more Pytorch-Lightning examples.

1. Run wandb sweep sweep.yaml

2. Run wandb agent where is given by previous command

3. Visualize and compare the sweep runs

   

Results

After running the script a few times, you will be able to compare quickly a large combination of hyperparameters.

Feel free to modify the script and define your own hyperparameters.

See the live report →