Research map / Detection and segmentation / Open-set and few-shot detection
Kiroshi: An Agentic Perception System for High-Accuracy Image Parsing
Haipeng ZHOU, Jinshan Liu, He Zhang, Xuequan Lu, Jun Ma, Lei Zhu
Read the paperCheck your idea against 137,217 papers
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 DPO-aligned MLLM policy with 3 refinement rounds vs vanilla box (ablation) | Kiroshi · HiFiRefMS segmentation | maxF_beta: 0.764 -> 0.875 | Improved |
| add DPO-aligned MLLM policy with 3 refinement rounds vs vanilla box (ablation) | Kiroshi · HiFiRefMS matting | SAD: 32.16 -> 26.34 | Improved |
| increase refinement rounds from 3 to 5 (ablation) | Kiroshi · HiFiRefMS segmentation | maxF_beta: 0.875 -> 0.863 | Hurt |
| increase refinement rounds from 1 to 3 (ablation) | Kiroshi · HiFiRefMS segmentation | HCE: 1514 -> 1218 | Improved |
| use 32x32 grid instead of 64x64 (ablation) | Kiroshi · HiFiRefMS segmentation | maxF_beta: 0.820 -> 0.875 | Improved |
| use 32x32 grid instead of 16x16 (ablation) | Kiroshi · HiFiRefMS segmentation | maxF_beta: 0.862 -> 0.875 | Improved |
| use Ovis2.5-2B instead of Qwen2.5-VL-3B backbone (ablation) | Kiroshi · HiFiRefMS segmentation | maxF_beta: 0.852 -> 0.875 | Improved |
| use Qwen3-VL-2B instead of Ovis2.5-2B backbone (ablation) | Kiroshi · HiFiRefMS matting | SAD: 26.34 -> 24.37 | Mixed |
| use Kiroshi instead of ZIM | Kiroshi · HiFiRefMS matting | SAD: 36.41 -> 26.34 | Improved |
| run mask-guided refinement x3 with external pipelines instead of x1 | Text4Seg + MGMatting · HiFiRefMS segmentation | maxF_beta: 0.846 -> 0.833 | Hurt |
| run Kiroshi refinement x3 instead of x1 | Kiroshi · HiFiRefMS segmentation | maxF_beta: 0.838 -> 0.875 | Improved |
Closest papers by meaning
- Revisiting Context Aggregation for Image Matting (ICML 2024 · similarity 0.86)
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.85)
- PixelCraft: A Multi-Agent system for High-Fidelity Visual Reasoning on Structured Images (ICLR 2026 · similarity 0.85)
- Privileged Prior Information Distillation for Image Matting (AAAI 2024 · similarity 0.85)
- In-Context Matting (CVPR 2024 · similarity 0.85)
- Training Matting Models Without Alpha Labels (AAAI 2025 · similarity 0.85)
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.
