NVIDIA Releases c-foundationstereo-s Depth Estimation Model on HuggingFace

NVIDIA has published the c-foundationstereo-s model on HuggingFace for the depth-estimation task. This is FoundationStereo — a model for estimating depth from a pair of RGB images from a stereo camera; it outputs a disparity map. The developer positions it as a solution with strong zero-shot generalization for industry, robotics and smart spaces.

What it means

The largest weights file is 0.8 GB in ONNX format, the repository totals 4.1 GB, and there are also weights in PTH format. The model runs via TensorRT. License — NVIDIA Open Model License: commercial use is permitted, you may create and distribute derivative models, but when distributing you must include the license text and a Notice file with attribution. Use must comply with NVIDIA's Trustworthy AI terms; circumventing the model's technical limitations and safeguards is prohibited. The developer does not specify exact VRAM requirements, but the 0.8 GB weight of the main file and the focus on TensorRT indicate that the model is designed to run on local hardware with an NVIDIA GPU. You can follow the release of similar models in the CPU3D news feed.