RTMPose-tiny — acaua mirror (pure-PyTorch port)
=================================================

This product includes:

1. PORTED SOURCE CODE: a pure-PyTorch port of the RTMPose architecture
   (located in the acaua repository at src/acaua/adapters/rtmpose/) is a
   derivative work of:

     - OpenMMLab's mmpose implementation
       https://github.com/open-mmlab/mmpose @ commit
       759b39c13fea6ba094afc1fa932f51dc1b11cbf9 — Apache-2.0
       Files derived from:
         mmpose/models/heads/coord_cls_heads/rtmcc_head.py  (RTMCCHead)
         mmpose/models/utils/rtmcc_block.py                 (RTMCCBlock, Scale)
         mmpose/models/utils/transformer.py                 (ScaleNorm)
         mmpose/codecs/utils/post_processing.py             (get_simcc_maximum)
         mmpose/codecs/simcc_label.py                       (SimCCLabel.decode)
         mmpose/datasets/transforms/topdown_transforms.py   (TopdownAffine)
         mmpose/structures/bbox/transforms.py               (get_warp_matrix,
                                                            _rotate_point,
                                                            _get_3rd_point)

     - OpenMMLab's mmdetection implementation (CSPNeXt backbone)
       https://github.com/open-mmlab/mmdetection @ commit
       cfd5d3a985b0249de009b67d04f37263e11cdf3d — Apache-2.0
       Files derived from:
         mmdet/models/backbones/cspnext.py
         mmdet/models/backbones/csp_darknet.py    (SPPBottleneck)
         mmdet/models/layers/csp_layer.py         (CSPLayer, CSPNeXtBlock)
         mmdet/models/layers/se_layer.py          (ChannelAttention)

2. CONVERTED WEIGHTS: the model.safetensors file in this mirror is a
   key-cleaned conversion of the upstream pretrained checkpoint:

     - upstream URL:    https://download.openmmlab.com/mmpose/v1/projects/
                        rtmposev1/rtmpose-tiny_simcc-aic-coco_pt-aic-coco_
                        420e-256x192-cfc8f33d_20230126.pth
     - upstream SHA256: e84eb5b9ee9432259bdd19d6a01156604ba27139ca6373ddb4
                        ee7aa290d528e9
     - upstream paper:  Jiang et al., "RTMPose: Real-Time Multi-Person Pose
                        Estimation based on MMPose", arXiv:2303.07399
     - training data:   AI Challenger (AIC) + COCO
     - eval dataset:    COCO 17-keypoint (val 68.5 AP)

   Conversion was performed by scripts/convert_rtmpose.py in the acaua
   repository. The conversion only strips training-only buffers
   (data_preprocessor.*); no parameter renames or reshapes are applied,
   so the conversion is reversible and lossless.

3. BUNDLED DETECTOR: the acaua RTMPoseAdapter at runtime composes with
   CondadosAI/rtmdet_t_coco (a separate mirror with its own NOTICE and
   provenance) to produce a top-down pose pipeline. The detector mirror
   is a pure-PyTorch port of RTMDet-tiny under the same Apache-2.0
   license; see https://huggingface.co/CondadosAI/rtmdet_t_coco/blob/main/
   NOTICE for that mirror's full attribution chain.

Mirrored on 2026-04-22 by CondadosAI.

License
-------
Licensed under the Apache License, Version 2.0 (the "License"); you may
not use this file except in compliance with the License. You may obtain
a copy of the License at:

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.