Truck was the weakest-looking orbit in the gallery. Of ten orbit captures
baked the same way (64 evenly spaced inputs, long edge 960, DL3DV model),
bonsai scored highest on held-out views (27.2 dB vs 24.4 dB for Truck).
Baked with DVSM-TTB-p8-DL3DV-960p at 960x640; the thumbnail is frame 222.
The Space's 1 GB storage holds five examples, so Truck is removed.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Use front views of the subject for the example thumbnails
The thumbnails were the middle frame of each capture, which showed the
toy from the side, the statue from behind and the truck from the rear.
Pick a frame that faces the subject instead (toy frame 38 of the 64
extracted, DL3DV scene 9 frame 162, scene 139 frame 157, Truck frame 230).
mipnerf360_garden already faced the table and is unchanged.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The Space's 1 GB storage holds five examples; this one makes room for
iphone_toy. mipnerf360_garden stays.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Add an iPhone video example baked with the general model
A 29 s handheld orbit of a desk toy, reconstructed the way an upload is:
64 evenly spaced frames, pi3x poses, DVSM-TTB-p8-general-960p at 960x544.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The two DL3DV gallery scenes were baked from the distorted captures with
the principal point assumed at the image centre. Rebake them from the
undistorted DL3DV-140 frames with the corrected intrinsics; same 64 views,
960x544 and model as before. Thumbnails now use the undistorted frame too.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
