AI GitHub
MMSegmentation for semantic segmentation experiments
Train and evaluate pixel-labeling models using shared configurations and dataset interfaces.
Key Takeaways
- MMSegmentation assembles the architecture, transforms, dataset and optimizer through configuration.
- Prepare a matching dataset configuration and model dependencies.
- Reproduce one documented configuration on a small subset and visualize predictions alongside labels.
- Library versions and pretrained checkpoints need to agree.
How it works
MMSegmentation assembles the architecture, transforms, dataset and optimizer through configuration. A common interface makes controlled comparisons possible across models.
What you need
Prepare a matching dataset configuration and model dependencies. Confirm label mapping and ignore values before starting a run, since they affect metrics and visualizations.
What to try first
Reproduce one documented configuration on a small subset and visualize predictions alongside labels. Check the handling of ignored pixels and class indices.
Measure performance by class as well as overall. Check boundary errors on thin structures or small regions before concluding that the new model is better.
Where to look for failures
Library versions and pretrained checkpoints need to agree. Aggregate intersection-over-union can conceal weak performance on small or rare classes.
Related Franklin pages
Follow one annotation from labeling to training and evaluation. Check that frame numbers, class IDs and coordinate conventions survive the handoff.
- Explore images or video, annotations and model outputs together. FiftyOne for inspecting vision datasets and predictions
- Create image and video labeling tasks with review and export workflows. CVAT for computer vision annotation
For an overview of the types of predictions vision systems produce, see Franklin’s glossary. Computer vision
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗