BEiT v2: Masked Image Modeling with Vector-Quantized Visual Tokenizers

Official PyTorch implementation and pretrained models of BEiT v2.

The code and pretrained models of BEiT can be found at here.

The code and pretrained models of BEiT-3 can be found at here.

• March, 2023: release the code and pretrained models of BEiT-3
• March, 2023: BEiT-3 was accepted by CVPR 2023.
• Sept 2022: release the code and pretrained models of BEiT v2
• Aug 2022: release preprint Image as a Foreign Language: BEiT Pretraining for All Vision and Vision-Language Tasks
• Aug 2022: release preprint BEiT v2: Masked Image Modeling with Vector-Quantized Visual Tokenizers
• June 2022: release preprint VL-BEiT: Generative Vision-Language Pretraining
• March, 2022: add linear probe examples
• January, 2022: BEiT was accepted by ICLR 2022 as Oral presentation (54 out of 3391).
• August 2021: BEiT is on HuggingFace
• July 2021: BEiT-large achieves state-of-the-art results on ADE20K (a big jump to 57.0 mIoU) for semantic segmentation.
• July 2021: BEiT-large achieves state-of-the-art ImageNet top-1 accuracy (88.6%) under the setting without extra data other than ImageNet-22k.
• July 2021: release the code and pretrained models of BEiT
• June 2021: release preprint BEiT: BERT Pre-Training of Image Transformers

Pretrained Models

We provide four BEiT weights pretrained on ImageNet-1k. The models were pretrained with 224x224 resolution.

• BEiT-base: #layer=12; hidden=768; FFN factor=4x; #head=12; patch=16x16 (#parameters: 86M)
• BEiT-large: #layer=24; hidden=1024; FFN factor=4x; #head=16; patch=16x16 (#parameters: 304M)

Download checkpoints that are self-supervised pretrained on ImageNet-1k and then intermediate finetuned on ImageNet-21k (recommended):

• BEiT-base: beitv2_base_patch16_224_pt1k_ft21k
• BEiT-large: beitv2_large_patch16_224_pt1k_ft21k

Download checkpoints that are self-supervised pretrained on ImageNet-1k:

• BEiT-base: beitv2_base_patch16_224_pt1k
• BEiT-large: beitv2_large_patch16_224_pt1k

Setup

alias=`whoami | cut -d'.' -f2`; docker run -it --rm --runtime=nvidia --ipc=host --privileged -v /home/${alias}:/home/${alias} pytorch/pytorch:1.7.1-cuda11.0-cudnn8-devel bash

First, clone the repo and install required packages:

git clone https://github.com/microsoft/unilm.git
cd unilm/beit2
pip install -r requirements.txt

The required packages including: Pytorch version 1.7.1, torchvision version 0.8.2 and Timm version 0.4.12, etc.

For mixed-precision training, please install apex

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

Fine-tuning on ImageNet-1k (Image Classification)

We summarize the validation results as follows. We also provide the fine-tuned weights. The detailed instructions to reproduce the results can be found at get_started_for_image_classification.md.

name	initialized checkpoint	resolution	acc@1	acc@5	#params	weight
BEiTv2-base	beitv2_base_patch16_224_pt1k	224x224	85.5	97.5	86.5M	link
BEiTv2-base	beitv2_base_patch16_224_pt1k_ft21k	224x224	86.5	98.0	86.5M	link
BEiTv2-large	beitv2_base_patch16_224_pt1k	224x224	87.3	98.2	304M	link
BEiTv2-large	beitv2_base_patch16_224_pt1k_ft21k	224x224	88.4	98.6	304M	link

Fine-tuning on ADE20K (Semantic Segmentation)

We summarize the validation results as follows. We also provide the fine-tuned weights. The detailed instructions to reproduce the results can be found at semantic_segmentation/README.md.

name	initialized checkpoint	method	crop size	iterations	mIoU	#params	weight
BEiTv2-base	beitv2_base_patch16_224_pt1k	UPerNet	512x512	160k	53.1	163M	link
BEiTv2-base	beitv2_base_patch16_224_pt1k_ft21k	UPerNet	512x512	160k	53.5	163M	link
BEiTv2-large	beitv2_large_patch16_224_pt1k	UPerNet	512x512	160k	56.7	441M	link
BEiTv2-large	beitv2_large_patch16_224_pt1k_ft21k	UPerNet	512x512	160k	57.5	441M	link

Fine-tuning on MSCOCO2017 (Object Detection)

Under preparation.

Pre-training on ImageNet-1k

See PRETRAINING.md for detailed instructions.

Visual Tokenizer (VQ-KD) Trained on ImageNet-1k

We provide the VQ-KD tokenizer trained on ImageNet-1k.

• vqkd_encoder_base_decoder_3x768x12: #encoder layer=12; #decoder layer=3; hidden=768; FFN factor=4x; #head=12; patch=16x16;

See TOKENIZER.md for more details.

Citation

If you find this repository useful, please consider citing our work:

@inproceedings{beit,
title={{BEiT}: {BERT} Pre-Training of Image Transformers},
author={Hangbo Bao and Li Dong and Songhao Piao and Furu Wei},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=p-BhZSz59o4}
}

@article{beitv2,
title={{BEiT v2}: Masked Image Modeling with Vector-Quantized Visual Tokenizers},
author={Zhiliang Peng and Li Dong and Hangbo Bao and Qixiang Ye and Furu Wei},
year={2022},
eprint={2208.06366},
archivePrefix={arXiv},
primaryClass={cs.CV}
}

Acknowledgement

This repository is built using the BEiT, the CLIP, the DeiT, the Dino repository and the timm library.

License

This project is licensed under the license found in the LICENSE file in the root directory of this source tree.

Microsoft Open Source Code of Conduct

Contact Information

For help or issues using BEiT v2 models, please submit a GitHub issue.

For other communications, please contact Li Dong (lidong1@microsoft.com), Furu Wei (fuwei@microsoft.com).