Research map / Detection and segmentation / Open-set and few-shot detection
Universal Instance Perception As Object Discovery and Retrieval
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 early image-prompt fusion (Bi-XAtt) (ablation) | UNINEXT ResNet-50 · Ref-YouTube-VOS | J&F: +2.8 | Helped |
| add early image-prompt fusion (Bi-XAtt) (ablation) | UNINEXT ResNet-50 · YouTube-VOS 2018 | J&F: +21.4 | Helped |
| add early image-prompt fusion (Bi-XAtt) (ablation) | UNINEXT ResNet-50 · RefCOCO REC | P@0.5: +2.3 | Helped |
| add early image-prompt fusion (Bi-XAtt) (ablation) | UNINEXT ResNet-50 · COCO object detection | AP: +0.2 | No clear effect |
| add early image-prompt fusion (Bi-XAtt) (ablation) | UNINEXT ResNet-50 · YouTube-VIS 2019 | AP: +2.0 | Helped |
| use static learned object queries instead of prompt-conditioned dynamic queries (ablation) | UNINEXT ResNet-50 · YouTube-VIS 2019 | AP: +2.8 | Helped |
| use static learned object queries instead of prompt-conditioned dynamic queries (ablation) | UNINEXT ResNet-50 · COCO, RefCOCO, YouTube-VOS, Ref-YouTube-VOS | AP / P@0.5 / J&F: -0.6 AP, -0.1 P@0.5, -0.4 J&F, -0.4 J&F | Hurt |
| unified multi-task model instead of task-specific models (ablation) | UNINEXT ResNet-50 · five tasks (OD, REC, VOS, R-VOS, VIS) | task metrics: +0.5 AP (COCO), +2.1 (RefCOCO), +2.8 (YTBVOS), +4.0 (R-YTBVOS), +2.9 (VIS19) | Helped |
| joint unified training across tasks and datasets with prompt-based retrieval | UNINEXT · Object detection (COCO val2017) | box AP: +2.7 over DN-Deformable-DETR (ResNet-50) | Helped |
| ViT-H backbone with UNINEXT | UNINEXT · RES RefCOCOg val-u | oIoU: UNINEXT-H 74.67 vs LAVT 61.24 | Helped |
| unified model (instead of per-task) achieves best results across 20 benchmarks | UNINEXT · 20 benchmarks across 10 tasks | task metrics | Mixed |
Closest papers by meaning
- Aligning and Prompting Everything All at Once for Universal Visual Perception (CVPR 2024 · similarity 0.86)
- Perceive Anything: Recognize, Explain, Caption, and Segment Anything in Images and Videos (NeurIPS 2025 · similarity 0.86)
- Same or Not? Enhancing Visual Perception in Vision-Language Models (CVPR 2026 · similarity 0.84)
- UFO: A Unified Approach to Fine-grained Visual Perception via Open-ended Language Interface (NeurIPS 2025 · similarity 0.84)
- UniPixel: Unified Object Referring and Segmentation for Pixel-Level Visual Reasoning (NeurIPS 2025 · similarity 0.84)
- UniVS: Unified and Universal Video Segmentation with Prompts as Queries (CVPR 2024 · 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.