Summary of Cohd: a Counting-aware Hierarchical Decoding Framework For Generalized Referring Expression Segmentation, by Zhuoyan Luo et al.
CoHD: A Counting-Aware Hierarchical Decoding Framework for Generalized Referring Expression Segmentation
by Zhuoyan Luo, Yinghao Wu, Tianheng Cheng, Yong Liu, Yicheng Xiao, Hongfa Wang, Xiao-Ping Zhang, Yujiu Yang
First submitted to arxiv on: 24 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 A newly proposed Generalized Referring Expression Segmentation (GRES) method amplifies the classic RES formulation by incorporating complex scenarios involving multiple or non-target objects. Recent approaches addressed GRES by extending well-established RES frameworks with object-existence identification, but these methods struggle to represent comprehensive objects of varying granularity due to the encoding of multi-granularity object information into a single representation. Moreover, simple binary object-existence identification across all referent scenarios fails to account for inherent differences, leading to ambiguity in object understanding. To address these issues, researchers propose a Counting-Aware Hierarchical Decoding (CoHD) framework for GRES, which decouples referring semantics by granularity and aggregates it hierarchically with intra- and inter-selection. Additionally, the framework incorporates counting ability by embodying multiple/single/non-target scenarios into count- and category-level supervision, enabling comprehensive object perception. The proposed CoHD method outperforms state-of-the-art GRES methods on various benchmarks, including gRefCOCO, Ref-ZOM, R-RefCOCO, and RefCOCO. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The new Generalized Referring Expression Segmentation (GRES) helps computers better understand what objects are being referred to in sentences. Right now, computers have trouble understanding complex sentences that refer to multiple or non-target objects. Some researchers tried to fix this by adding a simple check to see if the object exists or not, but this didn’t work well because it didn’t account for differences between objects. To solve this problem, scientists proposed a new way of processing language called Counting-Aware Hierarchical Decoding (CoHD). This method breaks down complex sentences into smaller parts and then combines them in a special way to better understand what’s being referred to. The results show that CoHD works much better than previous methods on several tests. |
Keywords
» Artificial intelligence » Semantics
