| 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. |
|
|
