Source code for the method described in the paper "SynthSet: Generative Diffusion Model for Semantic Segmentation in Precision Agriculture".
This repository is a fork from the extendable ClasSeg (https://github.com/aheschl1/ClasSegPipeline) package, and implements functionality using the "super_resolution" and "unstable_diffusion" extensions. Please read the ClasSeg README for in depth details on how to use the pipeline.
This pipeline includes image-mask pair generation, as well as super resolution for doubling the resolution from
Super resolution is highly effective for maintining efficacy between images and masks.
This code defines the following pipeline, where the third row is acheived through setting mode to "concat" in the config. The bottom row is acheived through setting "gan_weight" > 0 in the config:
