Semantic Segmentation Comparison

This repository contains the codebase for inferring and evaluating different semantic segmentation models which follow the MMSegmentation scheme. This code was developed as part of the undergraduate thesis "Analysis of coastal images in turistic environments" presented at ULPGC.

HOWTO

1. Install MMSegmentation repository and its dependecies.

2. Write the config file:

   Expected JSON config:

   {
       dataset: {
           images folder: ...,
           ann folder: ...
       },
       models: [
           {
               name: ...,
               config: ...,
               checkpoint: ...,
               output dir: ...
           }, ...
       ]
   }
   

   With the original images folder and the annotations as grey images with values between 0 and 150.
   models should be a list with routes to the MMSegmentation-like config and checkpoints.

3. Run either compare_pipeline.py for default or compare_postprocess_pipeline.py to execute and compute the results.

TODO

• Adding postproccessings:
  1. Think about others

Authors

• Jonay Suárez Ramírez