Ming Dai, Wenxuan Cheng, Jiang-Jiang Liu, Sen yang, Wenxiao Cai, Yanpeng Sun, Wankou Yang
Referring Image Segmentation (RIS) is a challenging task that aims to segment objects in an image based on natural language expressions. While prior studies have predominantly concentrated on improving vision-language interactions and achieving fine-grained localization, a systematic analysis of the fundamental bottlenecks in existing RIS frameworks remains underexplored. To bridge this gap, we propose DeRIS, a novel framework that decomposes RIS into two key components: perception and cognition. This modular decomposition facilitates a systematic analysis of the primary bottlenecks impeding RIS performance. Our findings reveal that the predominant limitation lies not in perceptual deficiencies, but in the insufficient multi-modal cognitive capacity of current models. To mitigate this, we propose a Loopback Synergy mechanism, which enhances the synergy between the perception and cognition modules, thereby enabling precise segmentation while simultaneously improving robust image-text comprehension. Additionally, we analyze and introduce a simple non-referent sample conversion data augmentation to address the long-tail distribution issue related to target existence judgement in general scenarios. Notably, DeRIS demonstrates inherent adaptability to both non- and multi-referents scenarios without requiring specialized architectural modifications, enhancing its general applicability. The codes and models are available at https://github.com/Dmmm1997/DeRIS.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Instance Segmentation | RefCOCO testA | Mean IoU | 86.64 | DeRIS-L |
| Instance Segmentation | RefCOCO testA | Overall IoU | 86.49 | DeRIS-L |
| Instance Segmentation | RefCoCo val | Mean IoU | 85.72 | DeRIS-L |
| Instance Segmentation | RefCoCo val | Overall IoU | 85.41 | DeRIS-L |
| Instance Segmentation | RefCOCO testB | Mean IoU | 84.52 | DeRIS-L |
| Instance Segmentation | RefCOCO testB | Overall IoU | 82.87 | DeRIS-L |
| Instance Segmentation | RefCOCOg-test | Mean IoU | 81.32 | DeRIS-L |
| Instance Segmentation | RefCOCO+ val | Mean IoU | 81.28 | DeRIS-L |
| Instance Segmentation | RefCOCO+ val | Overall IoU | 79.01 | DeRIS-L |
| Instance Segmentation | RefCOCO+ test B | Mean IoU | 78.59 | DeRIS-L |
| Instance Segmentation | RefCOCO+ testA | Mean IoU | 83.74 | DeRIS-L |
| Instance Segmentation | RefCOCO+ testA | Overall IoU | 82.34 | DeRIS-L |
| Instance Segmentation | RefCOCOg-val | Mean IoU | 80.01 | DeRIS-L |
| Instance Segmentation | gRefCOCO | cIoU | 72 | DeRIS-L |
| Instance Segmentation | gRefCOCO | gIoU | 77.67 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO testA | Mean IoU | 86.64 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO testA | Overall IoU | 86.49 | DeRIS-L |
| Referring Expression Segmentation | RefCoCo val | Mean IoU | 85.72 | DeRIS-L |
| Referring Expression Segmentation | RefCoCo val | Overall IoU | 85.41 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO testB | Mean IoU | 84.52 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO testB | Overall IoU | 82.87 | DeRIS-L |
| Referring Expression Segmentation | RefCOCOg-test | Mean IoU | 81.32 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO+ val | Mean IoU | 81.28 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO+ val | Overall IoU | 79.01 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO+ test B | Mean IoU | 78.59 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO+ testA | Mean IoU | 83.74 | DeRIS-L |
| Referring Expression Segmentation | RefCOCO+ testA | Overall IoU | 82.34 | DeRIS-L |
| Referring Expression Segmentation | RefCOCOg-val | Mean IoU | 80.01 | DeRIS-L |
| Referring Expression Segmentation | gRefCOCO | cIoU | 72 | DeRIS-L |
| Referring Expression Segmentation | gRefCOCO | gIoU | 77.67 | DeRIS-L |