LibreDINO-DETR includes code ported from DINO:

Source: https://github.com/IDEA-Research/DINO
Commit: d84a491d41898b3befd8294d1cf2614661fc0953
License: Apache License 2.0.
Copyright (c) 2022 IDEA. All Rights Reserved.

The pinned DINO implementation carries notices for work derived from
Conditional DETR, DETR, and Deformable DETR:

Conditional DETR
Source: https://github.com/Atten4Vis/ConditionalDETR
License: Apache License 2.0.
Copyright (c) 2021 Microsoft. All Rights Reserved.

DETR
Source: https://github.com/facebookresearch/detr
License: Apache License 2.0.
Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.

Deformable DETR
Source: https://github.com/fundamentalvision/Deformable-DETR
License: Apache License 2.0.
Copyright (c) 2020 SenseTime. All Rights Reserved.

DINO also identifies DAB-DETR and DN-DETR as architectural predecessors;
their source repositories are released under Apache License 2.0:

Sources: https://github.com/IDEA-Research/DAB-DETR
         https://github.com/IDEA-Research/DN-DETR

The Swin-L backbone is ported through DINO's modified Swin Transformer
implementation:

Source: https://github.com/microsoft/Swin-Transformer
License: MIT.
Copyright (c) 2021 Microsoft.

LibreYOLO ports only the three released inference architectures. Training-only
denoising, matching, criteria, losses, data pipelines, and the compiled CUDA
multi-scale deformable-attention extension are not included. The ResNet tensor
container and pure-PyTorch deformable-attention core reuse LibreYOLO's existing
Deformable DETR port, which has its own NOTICE.
