Hi — really impressive work on Flash3D, thanks for open-sourcing it.
I'm integrating a Core ML conversion of the RealEstate10K-trained checkpoint into an indie macOS / visionOS app (single-image wigglegram renderer). Before I can ship that, I need to nail down the license terms for both the code and the weights — and right now the repo doesn't have a LICENSE file, and the checkpoint distributed via misc.download_pretrained_models doesn't come with license metadata either.
Would you be willing to declare:
- A license for the code in this repository (MIT, BSD-3-Clause, and Apache-2.0 are all fine from my side — your sibling project szymanowiczs/splatter-image is BSD-3-Clause, which would be a natural precedent).
- A license for the pretrained weights (the `re10k_v2.pth` checkpoint and any variants). If the weights need to be research-only because of a training-data constraint (RealEstate10K is research-license), that's important to know too — it would mean the converted Core ML package can't ship in a public build.
For context: I currently ship with Depth Anything V2 (Apache-2.0) as the default and keep Apple SHARP as a research-only engine. Flash3D would slot in between — clearly better than a depth-warp for disocclusions, without SHARP's research-only constraint if the licensing works out.
Happy to contribute a PR that adds the LICENSE file(s) once you pick terms.
Thanks!
Hi — really impressive work on Flash3D, thanks for open-sourcing it.
I'm integrating a Core ML conversion of the RealEstate10K-trained checkpoint into an indie macOS / visionOS app (single-image wigglegram renderer). Before I can ship that, I need to nail down the license terms for both the code and the weights — and right now the repo doesn't have a
LICENSEfile, and the checkpoint distributed viamisc.download_pretrained_modelsdoesn't come with license metadata either.Would you be willing to declare:
For context: I currently ship with Depth Anything V2 (Apache-2.0) as the default and keep Apple SHARP as a research-only engine. Flash3D would slot in between — clearly better than a depth-warp for disocclusions, without SHARP's research-only constraint if the licensing works out.
Happy to contribute a PR that adds the LICENSE file(s) once you pick terms.
Thanks!