| 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. |
|
|
