Amazing work, thank you for open-sourcing such a clean and useful codebase!
Could you kindly share:
- The testing scripts you used for segmentation experiments (or a brief guide on how to run evaluation).
- The exact mapping/rule set you applied to reduce Replica's 101 classes to the 83 used in your experiments (e.g., a JSON/YAML file or a short description of which categories were merged).
Did you use replica_color.yaml as the class-mapping file for the 83-class reduction, or did you create a custom mapping?
Having these details would make it much easier to reproduce your results and build on top of them.
Thanks again for the great repo!