Research map / Detection and segmentation / CNN features
NICE: Improving Panoptic Narrative Detection and Segmentation With Cascading Collaborative Learning
Haowei Wang, Jiayi Ji, Tianyu Guo, Yilong Yang, Xiaoshuai Sun, Rongrong Ji
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add box branch supervision and cascade (joint learning) to mask-only model (ablation) | NICE · PNG segmentation (PNS), All | average recall: 58.1 -> 62.3 | Improved |
| learn detection jointly with segmentation instead of box only (ablation) | NICE · PNG detection (PND), All | average recall: 43.2 -> 57.7 | Improved |
| add Coordinate Guided Aggregation (CGA) (ablation) | NICE-dagger · PNG segmentation (PNS), All | average recall: 58.6 -> 60.4 | Improved |
| add Barycenter Driven Localization (BDL) after CGA (ablation) | NICE · PNG detection (PND), All | average recall: 45.9 -> 57.7 | Improved |
| use cascade instead of two parallel branches (ablation) | NICE · PNG detection (PND), All | average recall: 45.4 -> 57.7 | Improved |
| use Average of mask points as barycenter instead of Top1 (ablation) | BDL · PNG detection (PND), All | average recall: 30.4 -> 57.7 | Improved |
| use multi-scale features in BDL instead of fused features (ablation) | BDL · PNG detection (PND) vs segmentation | average recall: 58.7 vs 57.7 detection; 60.5 vs 62.3 segmentation | Mixed |
| include stuff phrases (ablation) | NICE · PNG detection (PND), All | average recall: 48.9 -> 57.7 | Improved |
| predict box with BDL instead of enclosing the mask (ablation) | NICE · PNG detection (PND), All | average recall: 47.5 -> 57.7 | Improved |
| use cascade design instead of two-branch to reduce prediction conflict (ablation) | NICE · PNG inconsistency error | IE: 27.2% -> 18.3% | Improved |
| train NICE on boxes only for visual grounding | NICE · Flickr30K | AP50: 67.08 -> 68.71 | Improved |
| zero-shot transfer instead of random predictions | NICE · RefCOCO testA RES | AP@0.5: 0.2 -> 34.8 | Improved |
Closest papers by meaning
- Towards Real-Time Panoptic Narrative Grounding by an End-to-End Grounding Network (AAAI 2023 · similarity 0.89)
- Improving Panoptic Narrative Grounding by Harnessing Semantic Relationships and Visual Confirmation (AAAI 2024 · similarity 0.88)
- Panoptic Narrative Grounding (ICCV 2021 · similarity 0.88)
- PiGLET: Pixel-Level Grounding of Language Expressions With Transformers (TPAMI 2023 · similarity 0.86)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.85)
- COCONut-PanCap: Joint Panoptic Segmentation and Grounded Captions for Fine-Grained Understanding and Generation (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
