Hypersim: Results as Published

Hypersim is a photorealistic synthetic dataset for indoor scene understanding, not a question-answering test. It has 77,400 images of 461 indoor scenes, with per-pixel depth, surface normals, instance labels, camera information, and a split of each image into reflectance, illumination, and a residual lighting term. The paper uses the images as pre-training and then scores transfer to semantic segmentation and to 3D shape prediction. It does not define the single unnamed number some model cards print in a Hypersim cell.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.