| ---
|
| license: mit
|
| library_name: libreyolo
|
| pipeline_tag: zero-shot-image-classification
|
| tags:
|
| - clip
|
| - zero-shot-image-classification
|
| - open-vocabulary
|
| - libreyolo
|
| ---
|
|
|
| # LibreCLIPb32-cls
|
|
|
| OpenCLIP ViT-B/32 (LAION-2B), repackaged as a native LibreYOLO checkpoint for
|
| **zero-shot, open-vocabulary** image classification with `LibreCLIP`. No
|
| training and no fixed label set: call `set_classes([...])`, then predict.
|
|
|
| ## Source
|
|
|
| Derived from [laion/CLIP-ViT-B-32-laion2B-s34B-b79K](https://huggingface.co/laion/CLIP-ViT-B-32-laion2B-s34B-b79K)
|
| (OpenCLIP arch `ViT-B-32`, pretrained tag `laion2b_s34b_b79k`).
|
| Copyright (c) 2021 OpenAI; (c) 2012-2021 OpenCLIP authors. Licensed under the
|
| MIT License.
|
|
|
| ## Data provenance
|
|
|
| These weights were trained on LAION-2B, which has a documented CSAM-content
|
| history (Stanford, December 2023); LAION subsequently released the cleaned
|
| Re-LAION. Prefer Re-LAION-derived weights where available. See `NOTICE`.
|
|
|
| ## Modifications
|
|
|
| State-dict key remapping only — LibreCLIP's native towers mirror the OpenCLIP
|
| module structure, so the load is 0-missing / 0-unexpected. Learned parameters
|
| are unchanged. See `weights/convert_clip_weights.py` in the
|
| [LibreYOLO source repository](https://github.com/LibreYOLO/libreyolo).
|
|
|
| ## Usage
|
|
|
| ```python
|
| from libreyolo import LibreCLIP
|
|
|
| model = LibreCLIP("LibreCLIPb32-cls.pt") # autodownloads from this repo
|
| model.set_classes(["a forklift", "an empty aisle", "a spill"])
|
| r = model.predict("warehouse.jpg")[0]
|
| print(model.names[r.probs.top1], float(r.probs.top1conf))
|
| ```
|
|
|
| ## License
|
|
|
| MIT License. See the [`LICENSE`](./LICENSE) and [`NOTICE`](./NOTICE) files.
|
|
|
