Train one model for three image segmentation tasks.
Trains, evaluates, quantizes, and exports OneFormer segmentation models and runs them on images.
When to use it
Use for OneFormer training, evaluation, export, quantization, or inference, including requests for universal or task-conditioned segmentation.
Give it segmentation datasets and any checkpoint required by the requested operation; it produces model results in a results directory.
What you provide
Requires Docker for the container-based workflow.
Requires NVIDIA Container Toolkit alongside Docker.
States a minimum of two GPUs with at least 24 GB VRAM per GPU; recommends four GPUs and A100 hardware.
This skill
GitHub
Downloads pretrained Swin weights