To provide participants with practical experience in setting up, configuring, and running semantic segmentation tasks using the mmSegmentation framework.
Breakdown
Exploring Configurations
Navigating the configuration system in mmSegmentation.
Modifying a sample config: Setting dataset paths, model type (e.g., U-Net), hyperparameters, etc.
Training a Model
Launching a training session using a sample configuration.
Monitoring training progress: Observing loss values, potential visualizations, etc.
Tips: Highlighting the importance of checkpoints and logging.
Evaluating the Model
Using mmSegmentation tools to evaluate the trained model's performance.
Interpreting common metrics for semantic segmentation (e.g., mIoU).
Visualizing segmentation results on sample images.
Fine-tuning with a Pre-trained Model
Loading a pre-trained model from the mmSegmentation model zoo.
Fine-tuning it on the RTS data subset or another sample dataset.
Observing the benefits of transfer learning in practice.
Discussion and Troubleshooting
Sharing insights or observations from the exercises.
Encouraging participants to discuss their experiences and any modifications they tried.
MMSegmentation Hands-on Lab
Goal
To provide participants with practical experience in setting up, configuring, and running semantic segmentation tasks using the mmSegmentation framework.
Breakdown