| 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. |
|
|
