Webssl Mae1b Full2b 224

--- library_name: transformers license: cc-by-nc-4.0 inference: false --- # Web-SSL MAE ViT-1B: 2B MetaCLIP data, 224 Resolution A 1 billion parameter Vision Transformer (ViT) trained with Masked Autoencoder (MAE) self-supervised learning on web-scale image data without language supervision. Introduced in ["Scaling Language-Free Visual Representation Learning"](https://arxiv.org/abs/2504.01017) (Fan et al., 2025). ## Model Details - **Architecture**: ViT (1536 width, 40 depth, 24 heads) - **Parameters**: 1B - **Resolution**: 224×224 pixels - **Training**: Self-supervised Web-MAE on 2B image samples from MetaCLIP web data ## Model Descriptions Web-SSL MAE 1B is a 1 billion parameter Vision Transformer model trained using masked autoencoder self-supervised learning on 2 billion web images without language supervision. This model demonstrates that pure visual learning, when scaled appropriately, can match or exceed the performance of language-supervised models like CLIP across various vision tasks. Web-MAE exhibits particularly strong performance on OCR & Chart understanding tasks while maintaining competitive performance across traditional vision benchmarks and multimodal tasks. <img src="webssl_teaser.png" alt="WebSSL Model Overview" width="600"> ## Usage ```python from transformers import AutoImageProcessor, ViTModel import torch from PIL import Image # Adjust the size, crop_size, etc. fields to your liking processor = AutoImageProcessor.from_pretrained('facebook/webssl-mae1b-full2b-224') model = ViTModel.from_pretrained('facebook/webssl-mae1b-full2b-224').cuda().eval() # Process an image image = Image.open('path/to/image.jpg') inputs = processor(images=image, return_tensors="pt").to('cuda') with torch.no_grad(): outputs = model(**inputs) # Extract features from the encoder encoder_hidden_states = outputs.last_hidden_state ``` ## Citation ```bibtex @article{fan2025scaling, title={Scaling Language-Free Visual Representation Learning}, author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie}, year={2025}, eprint={2504.01017}, archivePrefix={arXiv}, primaryClass={cs.CV} } ```