License of two exact RTMPose-project checkpoints for proprietary server-side SaaS
We use MMPose/MMDetection code under Apache-2.0. This question is specifically about the trained weights.
We will not assume that the repository Apache-2.0 license covers these checkpoint files unless OpenMMLab confirms that it does.
Exact detector
rtmdet_nano_8xb32-100e_coco-obj365-person-05d8511e.pth
SHA-256:
05d8511e7b3fabc62e27d2f624179e004ad14ee63a86ca9d9d22c88f3db0eee1
Exact pose
rtmpose-s_simcc-coco_pt-aic-coco_420e-256x192-8edcf0d7_20230127.pth
SHA-256:
29dcacbb5c5f3ab2f03a67fedcb58cf7287f93b9a8a9d893f42416b63fc304ba
Intended use
Proprietary commercial SaaS:
- server-side inference
- weights stay on the server
- customers do not download the weights
Please answer per file (yes / no / unknown) if you are able.
What license, if any, applies to this exact checkpoint? Apache-2.0? another license? no separately published license?
May this exact checkpoint be used for inference in a proprietary commercial SaaS where:
- weights remain server-side
- customers never receive the weights
Does the answer change if the weight is converted to ONNX or TensorRT but remains server-side?
May the raw
.pthcheckpoint be redistributed?May converted ONNX / TensorRT weights be redistributed?
May the checkpoint be fine-tuned and the derivative used commercially?
For RTMDet: was Objects365 actually used to train
05d8511e? If yes: do the Objects365 terms impose any restriction on commercial inference using this checkpoint?For RTMPose file
8edcf0d7, the config initializes fromcspnext-s_udp-aic-coco_.... Please confirm:- whether AIC appears only in backbone pretraining
- whether AIC is also involved elsewhere in this checkpoint's training lineage
- whether AIC terms impose restrictions on commercial inference
What attribution / NOTICE / license files are required if use is permitted?
Is there an official document that states the license applicable to published OpenMMLab
.pthmodel zoo checkpoints?
A link to an official weight-license document would close this issue.
Thank you.
Source: open-mmlab/mmpose