UniFormer-S (Kinetics-400) - acaua mirror (pure-PyTorch port)
=============================================================

This product includes:

1. PORTED SOURCE CODE: a pure-PyTorch port of the UniFormer-3D architecture
   (located in the acaua repository at src/acaua/adapters/uniformer/) is a
   derivative work of:

     - Sense-X/UniFormer (Apache-2.0)
       https://github.com/Sense-X/UniFormer @ main
       Files derived from:
         video_classification/models/uniformer.py
         (CBlock / SABlock / Attention / PatchEmbed / SpecialPatchEmbed /
          conv_3xnxn / conv_1xnxn helpers and the four-stage 3D assembly)

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

     - upstream HF mirror: Sense-X/uniformer_video (MIT)
     - upstream revision:  f9448914e6161573b14ba47b72fcef170e03a1f9
     - upstream filename:  uniformer_small_k400_16x8.pth
     - upstream SHA256:    d5fd7b0c49ee6a5422ef5d0c884d962c742003bfbd900747485eb99fa269d0db
     - upstream paper:     Li et al., "UniFormer: Unified Transformer for
                           Efficient Spatiotemporal Representation Learning",
                           arXiv:2201.04676.

   Key remap rules (eager, at conversion time):
       patch_embed*  ->  backbone.patch_embed*
       blocks*       ->  backbone.blocks*
       norm.*        ->  backbone.norm.*
       head.*        ->  head.fc.*

   Conversion was performed by scripts/convert_uniformer_video.py in the
   acaua repository. The conversion is reversible (state dict keys can be
   remapped back) and lossless (no quantization or pruning). The adapter
   loads the remapped state dict under load_state_dict(strict=True).

3. KINETICS-400 LABELS: labels.json is a JSON array of 400 action labels
   derived from the upstream kinetics_class_index.py file at the same
   Sense-X/uniformer_video revision (MIT). The Kinetics dataset itself
   (video content on YouTube) is NOT redistributed here - only the label
   vocabulary is.

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.