UniFormer-S_h14 COCO-pretrained backbone - acaua mirror
========================================================

This product includes:

1. PORTED SOURCE CODE: a pure-PyTorch port of the UniFormer dense-
   prediction backbone (located in the acaua repository at
   src/acaua/adapters/uniformer/_backbone_dense.py and _blocks.py) is a
   derivative work of:

     - Sense-X/UniFormer (Apache-2.0)
       https://github.com/Sense-X/UniFormer @ main
       Files derived from:
         object_detection/mmdet/models/backbones/uniformer.py
         (CBlock / SABlock / Attention / PatchEmbed /
          SABlock_Windows, window_partition, window_reverse,
          and the 4-stage dense-prediction assembly)

2. CONVERTED WEIGHTS: the model.safetensors file in this mirror is the
   backbone-only sub-tree of the upstream full Mask R-CNN checkpoint:

     - upstream repo:      Sense-X/UniFormer (Apache-2.0)
     - upstream weights:   Google Drive file id
                           13KhBYkHKQg-CyhAgn1LQM1K0R4bwSpWT
     - upstream filename:  mask_rcnn_1x_uniformer_s_h14.pth
     - upstream SHA256:    aa1e6bbec1c83344de96705f0e1aee853f1eec78df365e41ec802c202f00d9cf
     - upstream paper:     Li et al., "UniFormer: Unifying Convolution
                           and Self-attention for Visual Recognition",
                           arXiv:2201.09450 (ICLR 2022)

   Head extraction: the full Mask R-CNN checkpoint has 374 state-dict
   keys (backbone + FPN + RPN + ROI bbox head + ROI mask head). This
   mirror keeps only the 332 backbone.* keys, with the "backbone."
   prefix stripped. The discarded prefixes are:
         neck., rpn_head., roi_head., decode_head., auxiliary_head.

   Conversion was performed by scripts/convert_uniformer_dense_backbone.py
   in the acaua repository. The adapter loads the safetensors file
   under load_state_dict(strict=True) — key-naming drift between
   upstream and the port is a hard error at load time.

Mirrored on 2026-04-24 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.