The repository provides code for running inference with the Meta Segment Anything Model 2 (SAM 2), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.
This PR introduces a SAM-2.1 fine-tuning notebook. The notebook uses Roboflow to generate a dataset in the SAM-2.1 mask format for use in training models with image datasets.
The notebook is added to the notebooks directory, and written for use in Google Colab.
The notebook walks through an example of training on a car parts dataset. The dataset includes refined annotations that point to specific parts of a car, such as individual windows and doors. Out of the box, SAM-2.1 tends to label larger regions (i.e. the whole chassis, or the whole car), but with fine-tuning the model starts to improve at more precisely segmenting parts.
The notebook has been tested twice to ensure the code works as expected.
The Roboflow team has published an accompanying tutorial that walks through the notebook in more depth.
This PR introduces a SAM-2.1 fine-tuning notebook. The notebook uses Roboflow to generate a dataset in the SAM-2.1 mask format for use in training models with image datasets.
The notebook is added to the
notebooksdirectory, and written for use in Google Colab.The notebook walks through an example of training on a car parts dataset. The dataset includes refined annotations that point to specific parts of a car, such as individual windows and doors. Out of the box, SAM-2.1 tends to label larger regions (i.e. the whole chassis, or the whole car), but with fine-tuning the model starts to improve at more precisely segmenting parts.
The notebook has been tested twice to ensure the code works as expected.
The Roboflow team has published an accompanying tutorial that walks through the notebook in more depth.