Summary of Mapper: Multimodal Prior-guided Parameter Efficient Tuning For Referring Expression Comprehension, by Ting Liu et al.
MaPPER: Multimodal Prior-guided Parameter Efficient Tuning for Referring Expression Comprehension
by Ting Liu, Zunnan Xu, Yue Hu, Liangtao Shi, Zhiqiang Wang, Quanjun Yin
First submitted to arxiv on: 20 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper addresses the task of Referring Expression Comprehension (REC), which involves grounding a local visual region to natural language. The existing methods rely on multimodal alignment, leveraging powerful pre-trained models through full fine-tuning. However, this approach can break prior knowledge and incur significant computational costs. To solve REC in an efficient manner, the authors propose a novel framework called MaPPER, which incorporates Dynamic Prior Adapters, Local Convolution Adapters, and a Prior-Guided Text module. MaPPER achieves state-of-the-art results on three benchmarks with only 1.41% tunable backbone parameters, outperforming full fine-tuning and other Parameter-Efficient Transfer Learning (PETL) methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper aims to improve Referring Expression Comprehension by developing a more efficient approach. The current method uses pre-trained models and fine-tunes the entire backbone, but this can be slow and might lose the knowledge gained during training. To solve this problem, the authors create a new way of learning called MaPPER. This method combines different parts to help the model understand both visual and linguistic information. It does very well on three test sets and is better than other methods. |
Keywords
» Artificial intelligence » Alignment » Fine tuning » Grounding » Parameter efficient » Transfer learning
