Welcome to VLAA lab


Vision · Learning · Assured Autonomy

Our projects

Our lab has a broad interest in computer vision and machine learning, including developing efficient deep representation learning with minimal supervision and securing model performance under (adversarial) distribution shifts. In addition, we are interested in 1) building transparent and trustworthy medical AI systems that can function effectively in complex clinical environments; and 2) generative AI, including LLMs and Diffusion Models.

News

[June 2024]
  • Three papers are accepted by ECCV 2024.
  • We release Recap-DataComp-1B, where we use a LLaMA-3-powered LLaVA model to recaption the entire 1.3 billion images from DataComp-1B. Our Recap-DataComp-1B shows higher textual quality, and can help to train stronger CLIP models and T2I models.

    • [May 2024]
  • D-iGPT is accepted by ICML 2024. By scaling the model size to ViT-H, D-iGPT strongly secures an ImageNet top-1 accuracy of 90.0%.
  • Congratulations to Xianhang Li on winning the 2024-25 Jack Baskin & Peggy Downes-Baskin Fellowship.

    • [April 2024]
  • We release HQ-Edit, a dataset with high-resolution images & detailed and aligned editing instructions. Our fine-tuned InstructPix2Pix delivers superior editing performance.
  • Two papers are accepted by TMLR.

    • [March 2024]
  • One paper is accepted by NAACL 2024.
  • One paper is accepted by TMLR.

    • [February 2024]
  • Six papers are accepted by CVPR 2024.
  • One paper has been accepted by ISBI 2024.

    • [January 2024]
  • Tuning LayerNorm in Attention is accepted by ICLR 2024 as a spotlight paper.
  • Visual Probing is accepted by EACL 2024.

    • [December 2023]
  • Congratulations to Zijun and the team on winning the 2nd place in both the base model subtrack and the large model subtrack, Red Teaming Track, in NeurIPS 2023 Trojan Detection Challenge. We will release the code and the report soon.
  • We release D-iGPT, which attains 89.5% ImageNet top-1 accuracy with ViT-L. Larger models are coming!
  • One paper is accepted by TMLR.
  • Cihang Xie will serve as Area Chairs for ICML 2024.
  • Yuyin Zhou will serve as Area Chairs for CHIL 2024.

    • [November 2023]
  • CLIPA is accepted by NeurIPS 2023. In addition, we release our best model, CLIPA-G/14, which attains 83.0% zero-shot ImageNet top-1 accuracy.
  • CLIPA-v2 is accepted by NeurIPS 2023 R0-FoMo Workshop.
  • Sight Beyond Text is accepted by NeurIPS 2023 Instruction Workshop.

    • [October 2023]
  • Congratulations to Siwei and the team on winning the 2nd place in Task 4: Brain Metastases Segmentation of MICCAI 2023 BraTS Challenge.

    • [July 2023]
  • Cihang Xie and Yuyin Zhou will serve as Area Chairs for ICLR 2024.
  • Three papers (DiffMAE + SMAUG + BEVDistill) are accepted by ICCV 2023.

    • [June 2023]
  • Two papers (SwinMM + MLB-Seg) are accepted by MICCAI 2023.
  • Cihang Xie and Yuyin Zhou will serve as Area Chairs for CVPR 2024.
  • We release CLIPA, which enables CLIP training with limited computational resources. Furthermore, at a low cost of just $15,000, our CLIPA-v2 effectively elevates the zero-shot ImageNet top-1 accuracy to an impressive 81.8%.

    • [March 2023]
  • Yuyin Zhou receives the 2023 UCSC Hellman Fellowship.

    • [February 2023]
  • DMAE is accepted by CVPR 2023.
  • Cihang Xie will serve as an Area Chair for NeurIPS 2023.

    • [January 2023]
  • Two papers (RobustCNN + One-Pixel Shortcut) are accepted by ICLR 2023.
  • Yuyin Zhou will serve as an Area Chair for MICCAI 2023.

    • [December 2022]
  • One paper (SSL-FL) is accepted by IEEE TMI.
  • One paper (BNET) is accepted by IEEE TPAMI.
  • Yuyin Zhou will serve as an Area Chair for CHIL 2023.
  • Cihang Xie will serve as an Area Chair for ICML 2023.

    • [November 2022]
  • One paper (Practical Disruption of Deepfake) is accepted by AAAI 2023.

    • [October 2022]
  • Cihang Xie will be giving a talk in ECCV 2022 Workshop on Adversarial Robustness in the Real World.
  • Cihang Xie will serve as an Area Chair for CVPR 2023 and ICCV 2023.

    • [September 2022]
  • Three papers (Adversarial Attack on Attackers + Counterfactual Simulation Testing + Multi-Granularity Cross-modal Alignment) are accepted by NeurIPS 2022.
  • One paper (Across Feature Map Attention) is accepted by IEEE TPAMI.
  • Yuyin Zhou is selected as a finalist for MICCAI 2022 Young Scientist Publication Impact Award.
  • Cihang Xie will serve as an Area Chair for ICLR 2023.

    • [July 2022]
  • Two papers (Efficient Video Pretraining with Images + ViP) are accepted by ECCV 2022.
  • Two papers (CateNorm + MINiT) are accepted by MICCAI Workshop 2022.

    • [June 2022]
  • Our CVPR 2022 MCV workshop, co-organized by Yuyin Zhou, is featured in CVPR Daily and Best of CVPR!
  • Cihang Xie will be giving a talk in CVPR 2022 Workshop on The Art of Robustness: Devil and Angel in Adversarial Machine Learning.

    • [March 2022]
  • Four papers (Simulated Adversarial Testing + SDMP + CD2-pFed + CNN vs. Transformer in Federated Learning) are accepted by CVPR 2022.
  • Cihang Xie will serve as an Area Chair for NeurIPS 2022.

    • [February 2022]
  • Cihang Xie will be giving a talk in AAAI 2022 1st International Workshop on Practical Deep Learning in the Wild.
  • Cihang Xie will serve as an Area Chair for ECCV 2022.

    • [January 2022]
  • Two papers (Fast AdvProp + iBOT) are accepted by ICLR 2022.
  • Yuyin Zhou will serve as an Area Chair for MICCAI 2022.

    • [December 2021]
  • One paper (External Attention Assisted Training) is accepted by IEEE TMI.
  • Cihang Xie will be giving a talk in NeurIPS 2021 Visual Domain Adaptation Challenge.
  • Yuyin Zhou will serve as an Area Chair for CHIL 2022.

    • [October 2021]
  • Cihang Xie will be giving a talk in ICCV 2021 2nd Workshop on Adversarial Robustness In the Real World.
  • Our ICCV Computer Vision for Automated Medical Diagnosis Workshop, co-organized by Yuyin Zhou, is featured in ICCV Daily!

    • [September 2021]
  • One paper (CNN vs. Transformer) is accepted by NeurIPS 2021.

    • [July 2021]
  • One paper (CaliCO) is accepted by ICCV 2021.
  • Cihang Xie will be giving a talk in ICML 2021 Workshop on A Blessing in Disguise: The Prospects and Perils of Adversarial Machine Learning.

    • [June 2021]
  • One paper (Phenotype Embedding Model) is accepted by MICCAI 2021.
  • Yuyin Zhou will be giving a talk in CVPR 2021 Workshop on Medical Computer Vision.
  • Cihang Xie will be giving a talk in CVPR 2021 Tutorial on Adversarial Machine Learning in Computer Vision.
  • Cihang Xie will serve as an Area Chair for ICLR 2022.

    • [February 2021]
  • One paper (DetAdvProp) is accepted by CVPR 2021.
  • One paper (IAG-Net) is accepted by IEEE TMI.

    • [January 2021]
  • One paper (Shape-Texture Debiased Training) is accepted by ICLR 2021.
  • Cihang Xie will serve as an Area Chair for ICCV 2021.

    •