Benchmark

This is a release with 17 splits we organize from 10 different academic datasets as a benchmark to evaluate NVS models, spanning from single-view, sparse-view, and semi-dense-view regime. Check the benchmark folder for more details.

Assets for CLI Demo

This is a release with several example scenes if you want to use demo.py to interact with Seva. You can take reference from the data structures of these folders to work with your own data.

There are three scenes in total:

• dl3d140-165f5af8bfe32f70595a1c9393a6e442acf7af019998275144f605b89a306557 and re10k-000c3ab189999a83 are two scenes converted from two academic datasets, DL3DV-140 and RealEstate10K. They are suited for img2img and img2vid task.
• garden_flythrough is one scene with user-specified trajectory and groundtruth target views, similar to the usage case in the gradio demo. Dummy black images are created as placeholders. It is suited for img2trajvid task.