NVIDIA TAO Data Services
Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.

Licensed under the Apache License, Version 2.0 (see the LICENSE file in
the root of this repository or https://www.apache.org/licenses/LICENSE-2.0).

================================================================================
Third-Party Software Notices and Attributions
================================================================================

This product bundles, embeds, or is derived from third-party software. The
attribution notices below are provided in compliance with Section 4(d) of the
Apache License, Version 2.0, and with the redistribution/attribution terms of
the upstream licenses listed for each component.

Each third-party component remains subject to its own license. In the event of
any conflict, the third-party license terms govern that component's use,
reproduction, and distribution.

--------------------------------------------------------------------------------
1. PyTorch Image Models (timm)
--------------------------------------------------------------------------------

Origin:    https://github.com/rwightman/pytorch-image-models
Author:    Ross Wightman
License:   Apache License, Version 2.0
Used in:   nvidia_tao_ds/backbone/swin_utils.py
           nvidia_tao_ds/backbone/convnext_utils.py
           nvidia_tao_ds/backbone/fan.py
           nvidia_tao_ds/backbone/vision_transformer.py

Portions of the backbone implementations are copied or adapted from the timm
library:

    Copyright 2019-2022, Ross Wightman
    Licensed under the Apache License, Version 2.0.

--------------------------------------------------------------------------------
2. Swin Transformer
--------------------------------------------------------------------------------

Origin:    https://github.com/microsoft/Swin-Transformer
Authors:   Microsoft Corporation (written by Ze Liu)
License:   MIT License
Used in:   nvidia_tao_ds/backbone/swin_utils.py

Code and pretrained weight URLs are derived from the Microsoft Swin Transformer
project:

    Swin Transformer
    Copyright (c) 2021 Microsoft
    Permission is hereby granted, free of charge, to any person obtaining a copy
    of this software and associated documentation files (the "Software"), to
    deal in the Software without restriction, including without limitation the
    rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
    sell copies of the Software, and to permit persons to whom the Software is
    furnished to do so, subject to the following conditions:

    The above copyright notice and this permission notice shall be included in
    all copies or substantial portions of the Software.

    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
    IN THE SOFTWARE.

--------------------------------------------------------------------------------
3. ConvNeXt
--------------------------------------------------------------------------------

Origin:    https://github.com/facebookresearch/ConvNeXt
Authors:   Meta Platforms, Inc. and affiliates
License:   MIT License
Used in:   nvidia_tao_ds/backbone/convnext_utils.py

Portions of the ConvNeXt backbone implementation and pretrained weight URLs are
derived from Meta's ConvNeXt project:

    Copyright (c) Meta Platforms, Inc. and affiliates.
    Licensed under the MIT License.

--------------------------------------------------------------------------------
4. NVIDIA FAN (Fully Attentional Networks)
--------------------------------------------------------------------------------

Origin:    https://github.com/NVlabs/FAN
Authors:   NVIDIA Corporation, NVlabs
License:   NVIDIA Source Code License-NC
           (https://github.com/NVlabs/FAN/blob/main/LICENSE)
Used in:   nvidia_tao_ds/backbone/fan.py
           nvidia_tao_ds/backbone/swin_utils.py
           nvidia_tao_ds/backbone/convnext_utils.py

Portions are derived from the NVlabs FAN project. The FAN-derived portions are
provided under the NVIDIA Source Code License-NC; the surrounding NVIDIA
modifications are provided under Apache-2.0 as noted in the file headers. See
https://github.com/NVlabs/FAN/blob/main/LICENSE for the upstream terms.

--------------------------------------------------------------------------------
5. DINO (self-DIstillation with NO labels) / vision_transformer
--------------------------------------------------------------------------------

Origin:    https://github.com/facebookresearch/dino
Authors:   Meta Platforms, Inc. and affiliates
License:   Apache License, Version 2.0
Used in:   nvidia_tao_ds/backbone/vision_transformer.py

Portions of the Vision Transformer / positional-embedding interpolation code
are copy-pasted or adapted from Facebook's DINO project:

    Copyright 2021 Facebook, Inc. and its affiliates.
    Licensed under the Apache License, Version 2.0.

--------------------------------------------------------------------------------
6. XCiT (Cross-Covariance Image Transformers)
--------------------------------------------------------------------------------

Origin:    https://github.com/facebookresearch/xcit
Authors:   Meta Platforms, Inc. and affiliates
License:   Apache License, Version 2.0
Used in:   nvidia_tao_ds/backbone/swin_utils.py (positional encoding)

Portions of the positional-encoding implementation are based on the official
XCiT code:

    Copyright (c) Meta Platforms, Inc. and affiliates.
    Licensed under the Apache License, Version 2.0.

--------------------------------------------------------------------------------
7. Horovod (MPI rank/size environment helpers)
--------------------------------------------------------------------------------

Origin:    https://github.com/horovod/horovod
Authors:   Uber Technologies, Inc., The Linux Foundation
License:   Apache License, Version 2.0
Used in:   nvidia_tao_ds/augmentation/utils/distributed_utils.py

The `mpi_env_MPI_rank_and_size` helper is derived from Horovod's test utilities:

    Copyright 2017 Uber Technologies, Inc.
    Modifications copyright (C) 2019, NVIDIA CORPORATION.
    Licensed under the Apache License, Version 2.0.

--------------------------------------------------------------------------------
8. Grounding DINO
--------------------------------------------------------------------------------

Origin:    https://github.com/IDEA-Research/GroundingDINO
Authors:   IDEA-Research
License:   Apache License, Version 2.0
Used in:   nvidia_tao_ds/auto_label/grounding_dino/ (auto-label inference glue,
           tokenization helpers, plotting utilities). Core model code is
           consumed via the `nvidia_tao_pytorch` submodule, which carries its
           own NOTICE/LICENSE.

Portions of the autolabel pipeline integrate components derived from
Grounding DINO:

    Copyright 2023 IDEA-Research
    Licensed under the Apache License, Version 2.0.

--------------------------------------------------------------------------------
9. MAL (Mask Auto-Labeler)
--------------------------------------------------------------------------------

Origin:    https://github.com/NVlabs/mask-auto-labeler
Authors:   NVIDIA Corporation, NVlabs
License:   NVIDIA Source Code License-NC
Used in:   nvidia_tao_ds/auto_label/mal/ (inference glue). Core model code is
           consumed via the `nvidia_tao_pytorch` submodule, which carries its
           own NOTICE/LICENSE.

    Copyright (c) 2023, NVIDIA Corporation. All rights reserved.

================================================================================
Bundled Submodules
================================================================================

This repository declares the following git submodules in `.gitmodules`. They
are independently maintained NVIDIA repositories, distributed under their own
LICENSE/NOTICE files within each submodule's tree:

    tao-core      (NVIDIA TAO Toolkit core utilities)
    tao-pytorch   (NVIDIA TAO PyTorch model implementations)

Refer to the LICENSE and NOTICE files within each submodule, when initialized,
for attribution of the third-party software they vendor.

================================================================================
Bundled Python Wheels (CI / Release Artifacts)
================================================================================

The release/CI tooling under `Jenkinsfile.nightly` and `docker/` packages a set
of upstream Python wheels (declared in `docker/requirements-pip.txt`) into
release containers. Those wheels are not stored in this source repository, but
when redistributed as part of a release artifact they remain subject to their
upstream licenses. Representative components include (non-exhaustive):

    accelerate, cudf, cuml, h5py, lightning_fabric, mpi4py, opencv-python-headless,
    open-clip-torch, pycocotools, pyfacer, pytorch_lightning, scikit-image, timm,
    transformers, wandb, Cython, matplotlib, nvidia-dali, nvidia-nvjpeg2k, pandas,
    pyarrow, pytest, toml, fvcore, fairscale, torchmetrics, shapely, seaborn,
    google-genai, openai, apispec, apispec_webframeworks, flask, flask_limiter,
    marshmallow, marshmallow_enum, marshmallow-oneofschema, ruamel.yaml, orjson,
    pymongo, validators, timeloop, kubernetes, ngcsdk, leptonai, hydra-core,
    imagesize, omegaconf, prettytable, pynvml, ujson, pydocstyle, flake8,
    pyflakes, pylint, pytest-order, pyinstaller.

Each wheel is distributed under its own license. Their NOTICE files (where
present) are reproduced inside the wheels themselves and are surfaced with the
release container distribution; see the corresponding upstream projects for
authoritative attribution.

================================================================================
End of NOTICE
================================================================================
