OliverRensu / D-iGPT

[ICML 2024] This repository includes the official implementation of our paper "Rejuvenating image-GPT as Strong Visual Representation Learners"

D-iGPT

This repository is the official implementation of our Rejuvenating image-GPT as Strong Visual Representation Learners (accepted by ICML2024)

Introduction

This paper enhances image iGPT, one of the pioneering works that introduce autoregressive pretraining to predict next pixels for visual representation learning. Two simple yet essential changes are made. First, we shift the prediction target from raw pixels to semantic tokens, enabling a higher-level understanding of visual content. Second, we supplement the autoregressive modeling by instructing the model to predict not only the next tokens but also the visible tokens. This pipeline is particularly effective when semantic tokens are encoded by discriminatively trained models, such as CLIP. We introduce this novel approach as D-iGPT. Extensive experiments showcase that D-iGPT excels as a strong learner of visual representations: A notable achievement is its compelling performance on the ImageNet-1K dataset --- by training on publicly available datasets, D-iGPT unprecedentedly achieves 90.0% top-1 accuracy with a vanilla ViT-H. Additionally, D-iGPT shows strong generalization on the downstream task.

Model Training

We train our base size D-iGPT on torch+GPU and large/huge size D-iGPT on torch_xla+TPU. Please refer D-iGPT and D-iGPT_torchxla respectively.

Performance

Performance comparison on ImageNet-1K classification and ADE20K Semantic Segmentation.

Method	Model Size	Top-1	mIoU
MAE	ViT-B	83.6	48.1
RandSac	ViT-B	83.7	-
EVA	ViT-B	85.5	53.3
D-iGPT	ViT-B	86.2	53.8

The torch+GPU code produces better results. This is likely caused by the system difference between torch+GPU and torchxla+TPU.

	ViT-Base
torch+GPU	86.2
torchxla+TPU	85.9

Performance comparison on ImageNet-1K classification with IN-21K as training data (if avaiable).

Method	Model Size	Top-1
MAE	ViT-L	85.9
BEiT	ViT-L	88.6
D-iGPT	ViT-L	89.5
OpenClip	ViT-H	88.6
D-iGPT	ViT-H	90.0

Checkpoint

The pretrained models are available at [huggingface🤗]

Acknowledgement

We are very grateful that this work is supported by TPU Research Cloud (TRC) program and Google Cloud Research Credits program.

✍ Citation

@inproceedings{ren2024digpt,
   title = {Rejuvenating i-GPT for Scalable Visual Representation Learning},
   author = {Ren, Sucheng and Wang, Zeyu and Zhu, Hongru and Xiao, Junfei and Yuille, Alan and Xie, Cihang},
   booktitle = {ICML},
   year = {2024}
}

If you have any question, feel free to contact Sucheng Ren :)