A highly relevant CVPR paper on Automatic Self-Improved Data Engine using LLM and VLM

[wuyujack]

Hi all,

I want to bring the paper "AIDE: An Automatic Data Engine for Object Detection in Autonomous Driving" (CVPR 2024) to your attention as I believe it fully matches the spirit of LLM4Annotation. AIDE leverages Large Language Model (LLM) and Vision-Language Models (VLMs) to perform Issue Identification (i.e., what the task-oriented model should learn continually), Data Curation (i.e., retrieve the relevant raw data from large amount of unlabeled data pool), Auto-Labeling and Continual Training of the task-oriented model, Verification (i.e., using LLM to generate test cases to verify the continual trained model). We verify the scaling law of AIDE on Autonomous Driving, specifically on novel object detection. It would highly appreciated if you could include the AIDE paper in your survey. Thank you so much!

Paper Link: https://openaccess.thecvf.com/content/CVPR2024/html/Liang_AIDE_An_Automatic_Data_Engine_for_Object_Detection_in_Autonomous_CVPR_2024_paper.html

Best,
Mingfu Liang
Northwestern University

[Zhen-Tan-dmml]

Hi Mingfu,

Thanks for reaching out. We would be more than happy to include your paper. Can u please submit a PR from GitHub? We will update the manuscript later.

Thanks,
Zhen

[wuyujack]

Hi Mingfu,

Thanks for reaching out. We would be more than happy to include your paper. Can u please submit a PR from GitHub? We will update the manuscript later.

Thanks, Zhen

Hi Zhen,

The PR is committed. Thank you so much!

Best,
Mingfu