DA3-GIANT is a 1.15 billion parameter foundation model developed for multi-view depth estimation, camera pose estimation, and 3D Gaussian estimation. It forms part of the Depth Anything 3 series and employs a unified depth-ray representation along with support for pose conditioning.
The model provides capabilities in relative depth estimation, pose estimation, pose conditioning, and 3D Gaussians. It is positioned within computer vision tasks that include monocular depth and multi-view geometry.
Delivery occurs through the Hugging Face platform where the model card, safetensors files, and associated code are hosted. Installation requires cloning the GitHub repository from ByteDance-Seed, changing into the depth-anything-3 directory, and running pip install -e .. A basic Python example imports the DepthAnything3 class from the depth_anything_3.api module after installation. Command line interface tools are also included in the package.
The license is CC BY-NC 4.0, which restricts use to non-commercial purposes only. No pricing information is stated for the model itself.
DA3 GIANT is an Other AI product. It focuses on estimating accurate depth, camera poses, and 3D structure from multiple views in a unified model. DA3 GIANT is an open-source project aimed at computer vision researchers and developers. The project is open source (Apache-2.0). It runs on the web, the command line, and API.
ByteDance-Seed builds and maintains DA3 GIANT, and the product first shipped in 2025. The project is developed in the open on GitHub with 5.9k stars and 2 commits in the last 90 days. Among its 3 catalogued features are Depth Estimation, Pose Estimation, and 3D Gaussians.
Latest indexed changes and source events
depth-anything/DA3-GIANT verified by the PulseGate indexer
Other apps tracked under the same category.
