arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2402.05008cs.CVcs.AIcs.LG
收藏

EfficientViT-SAM:无精度损失的加速型Segment Anything Model

EfficientViT-SAM: Accelerated Segment Anything Model Without Accuracy Loss

Tsinghua University(清华大学);MIT(麻省理工学院);NVIDIA(英伟达)

Zhuoyang Zhang, Han Cai, Song Han

AI总结:

该研究提出加速版分割一切模型系列EfficientViT-SAM,用EfficientViT替换SAM的重型图像编码器,经知识蒸馏和SA-1B数据集端到端训练,在A100 GPU上实现48.9倍TensorRT速度提升且无精度损失。

摘要:

我们提出了EfficientViT-SAM,这是一个全新的加速型Segment Anything Model(SAM,分割一切模型)系列。我们保留了SAM的轻量级提示编码器与掩码解码器,同时用EfficientViT替换了原本沉重的图像编码器。训练阶段,我们首先通过SAM-ViT-H图像编码器向EfficientViT进行知识蒸馏,随后在SA-1B数据集上开展端到端训练。得益于EfficientViT的高效性与性能容量,在A100 GPU上经TensorRT实测,EfficientViT-SAM的速度是SAM-ViT-H的48.9倍,且未牺牲模型性能。我们的代码与预训练模型已发布于https://github.com/mit-han-lab/efficientvit。

英文摘要:

We present EfficientViT-SAM, a new family of accelerated segment anything models. We retain SAM's lightweight prompt encoder and mask decoder while replacing the heavy image encoder with EfficientViT. For the training, we begin with the knowledge distillation from the SAM-ViT-H image encoder to EfficientViT. Subsequently, we conduct end-to-end training on the SA-1B dataset. Benefiting from EfficientViT's efficiency and capacity, EfficientViT-SAM delivers 48.9x measured TensorRT speedup on A100 GPU over SAM-ViT-H without sacrificing performance. Our code and pre-trained models are released at https://github.com/mit-han-lab/efficientvit.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑