Licence of the pretrained weights (big-lama / lama-fourier) — clarification request

[hunainibnshakeel-crypto]

Hello, and thank you for releasing LaMa.

I'm evaluating LaMa for use in a commercial desktop application and I want to
get the licensing right rather than assume it. I could not find an answer in
the repository, so I'd be grateful for clarification.

The repository LICENSE is Apache-2.0 (Copyright 2021 Samsung Research).
My question is specifically about the pretrained weights (big-lama,
lama-fourier), which are distributed separately from the repository via
Google Drive rather than committed here:

1. Are the pretrained weights covered by the repository's Apache-2.0
   licence? I could not find a licence statement accompanying the weight
   downloads themselves.

2. If they are, does that grant extend to commercial use? The Apache-2.0
   text has no field-of-use restriction, but I want to confirm it was intended
   to cover the weights and not only the source code.

3. How should the training data terms be understood in relation to the
   weights? big-lama is described as trained on a subset of
   Places-Challenge, and the Places2 terms of use state the data may be used
   "only for non-commercial research and educational purposes". I am aware
   that CSAIL separately releases its own Places-trained CNNs under CC BY,
   which suggests the dataset terms are not understood to bind a trained
   model — but I would rather have your position than infer one.

For context on why I'm asking rather than relying on third parties: several
Hugging Face mirrors of these weights declare apache-2.0, but they are
re-uploads by parties who did not train the model (one states plainly that it
is "a port of the original PyTorch big-lama model"). A downstream
re-declaration isn't something I can treat as authoritative.

An explicit statement in the README or a LICENSE file alongside the weight
downloads would help everyone in this position, not just me.

Thank you for your time.

[skiy7j]

We plan to use the official big-lama pretrained checkpoint for commercial image inpainting in a private cloud service.
Could you please confirm whether the pretrained big-lama weights are licensed under Apache License 2.0, including commercial use and private cloud/container deployment?

[handoffpro]

Could a project maintainer confirm that the Big LaMa pretrained weights linked in your README (smartywu/big-lama) are covered by Apache 2.0 for use in a paid macOS app, including conversion to Core ML and redistribution of the converted weights, either bundled or as an optional download?
Do the Places training-data terms impose any additional restrictions on these uses? We would distribute only the converted model, not the training images.