pcuenq/marigold-coco-segmentation

CoCoGold: Object Segmentation with Marigold

This is a model trained with a repurposed version of Marigold that can do text-grounded object segmentation, i.e., you provide the name of the object you want to segment and the model will draw a segmentation mask on it.

The model was trained on the COCO dataset using this library to extract random crops and segmentation masks.

For more details, please refer to the post.

Checkpoints

• The main branch of the repo contains a diffusers pipeline with a UNet trained after ~18,000 steps.
• This branch contains intermediate checkpoints saved during the training run.