The HQ-SAM (High-Quality Segment Anything Model) is an upgrade to the Segment Anything Model (SAM) designed to accurately segment complex objects. It maintains SAM's zero-shot capabilities and promptable design while introducing minimal additional parameters and computation. The HQ-SAM uses a learnable High-Quality Output Token that is injected into SAM's mask decoder and fuses with early and final ViT features for improved mask details. It was trained on a dataset of 44K fine-grained masks and evaluated in a suite of 9 diverse segmentation datasets across different downstream tasks, including zero-shot transfer protocols. The code requires python>=3.8, pytorch>=1.7, and torchvision>=0.8 and can be installed through the provided instructions. Pretrained checkpoint models are available for download, and usage examples are provided in the demo and colab notebook. If used in research, the authors ask to cite their paper.

1.gif

sam-hq

The Segment Anything Model (SAM) is an object mask generation model that produces high-quality object masks from input prompts such as points or boxes, and it can be used to generate masks for all objects in an image. It has been trained on a dataset of 11 million images and 1.1 billion masks, and has strong zero-shot performance on a variety of segmentation tasks. The code requires Python 3.8 or higher, as well as PyTorch 1.7 and TorchVision 0.8, and can be installed via pip install or by cloning the repository locally. SAM's lightweight mask decoder can be exported to ONNX format and used in any environment that supports ONNX runtime. The code is licensed under Apache 2.0 and contributions are welcome.

model_diagram.png

Segment Anything Model (SAM)

A fast, easy way to create machine learning models for your sites, apps, and more – no expertise or coding required.

Teachable Machine.jpg

Teachable Machine

Roboflow is a computer vision company that helps innovators apply computer vision, train and deploy models, and conduct and manage experiments. You can easily upload, organize, and annotate your training data with dozens of annotation formats. Roboflow also provides a model library with pre-configured model architectures and custom Colab Notebooks. With Roboflow, you can host your trained model with a single click, deploy your model to the cloud, the edge or the browser, and customize your own models. Roboflow provides a streamlined workflow to identify edge cases and deploy fixes. Roboflow is trusted by over 250,000 developers building computer vision models.

Browse 4+ Computer Vision tools

sam-hq

Segment Anything Model (SAM)

Teachable Machine

Roboflow