Closed Jovp closed 3 years ago
Hi! Great question. You are correct that the meters_per_asset_unit
seems incorrect for this scene. The scale of each scene is specified by the artist, and it is possible for them to specify it incorrectly.
I won't go so far as to claim that every meters_per_asset_unit
value is exactly correct except for this one. But I did personally review every Hypersim scene during the development of the dataset, and ai_053_001
was the only scene that jumped out at me as having a noticeably incorrect scale. So I think our scale values are reliable, but imperfect.
I first noticed this particular scene when I was reviewing all of our camera trajectories as animations. In all the other scenes, the camera trajectories move around at qualitatively similar rates relative to the underlying scene (suggesting that our scale values are generally consistent across scenes). But in ai_053_001
specifically, the camera trajectory seemed like it was moving around much faster (suggesting a scale value that is inconsistent with the other scenes).
Also, minor typo correction, if meters_per_asset_unit
is 0.001, then each asset unit is 1mm 😅
Thanks a lot for your answers! Happy to learn that we are on the right path then!
Indeed this was a typo, I meant 0.01 :)
Hello,
For a research project, we are trying to use parts of the Hypersim dataset. We downloaded some metadata from some scenes and have access to the geometry. For scene 053_001, the meters_per_asset_unit is 0.001, meaning one asset unit should be 1cm. When looking at the scene and the camera positions this number seems quite off as it would mean the scene is roughly 1 meter large whereas it contains a full amphitheater. Can you confirm that this is an issue and that we are not missing something here? If this is an issue, how reliable do you think the other meters_per_asset_units should be?
Thanks a lot for providing this great dataset :)!
Best regards, Julien