Research map / Vision-language and multimodal / CLIP and zero-shot
PanoRec: Spatially-Structured Sequence Modeling for Multi-Granularity Panoramic Retrieval
Zidong Cao, Ding Zhou, Wenyao Gao, Lutao Jiang, Hui Xiong
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 MaxSim routing over cubemap faces instead of mean pooling (ablation) | PanoRec · ZInD (Full-FT, 8B) | Local Caption R@1: 16.25 -> 76.55 | Improved |
| add MaxSim routing over cubemap faces instead of attention pooling (ablation) | PanoRec · ZInD (Full-FT, 8B) | Local Caption R@1: 23.29 -> 76.55 | Improved |
| use multi-vector cubemap sequence instead of single-vector cubemap (ablation) | PanoRec · ZInD (Full-FT, 8B) | Local Caption R@1: 18.95 -> 74.78 | Improved |
| add downsampled global panorama to the sequence (ablation) | PanoRec · ZInD (Full-FT, 8B) | Global Caption R@1: 71.91 -> 79.68 | Improved |
| add downsampled global panorama to the sequence (ablation) | PanoRec · ZInD (Full-FT, 8B) | Local Image R@1: 82.46 -> 82.63 | No clear effect |
| switch single-vector cubemap input for ERP input (ablation) | ERP baseline · ZInD (Full-FT, 8B) | Local Caption R@1: 17.33 -> 18.95 | No clear effect |
| replace MaxSim with top-k averaging (k=2, k=4) (ablation) | PanoRec · ZInD (LoRA, 8B) | Local Caption R@1: 74.80 -> 48.03 (k=2), 21.38 (k=4) | Hurt |
| replace MaxSim with dynamic-temperature softmax weights (ablation) | PanoRec · ZInD (LoRA, 8B) | Local Caption R@1: 74.80 -> 75.29 | No clear effect |
| replace MaxSim with adaptive-threshold aggregation (ablation) | PanoRec · ZInD (LoRA, 8B) | Local Caption R@1: 74.80 -> 74.60 | No clear effect |
| add PanoRec sequence modeling over ERP baseline | Qwen3-VL-Embedding-8B · ZInD (LoRA) | Local Caption R@1: 11.50 -> 74.80 | Improved |
| add PanoRec sequence modeling over ERP baseline | Qwen3-VL-Embedding-8B · CVACT (Full-FT) | Global Caption R@1: 57.79 -> 61.91 | Improved |
| add PanoRec over direct training on panoramas | Qwen3-VL-8B · Matterport3D (zero-shot) | Local Caption R@1: 17.13 -> 82.78 | Improved |
Closest papers by meaning
- PanoWorld: Towards Spatial Supersensing in 360◦ Panorama World (NeurIPS 2026 · similarity 0.87)
- More than the Sum: Panorama-Language Models for Adverse Omni-Scenes (CVPR 2026 · similarity 0.87)
- Fine-Grained Perception in Panoramic Scenes: A Novel Task, Dataset, and Method for Object Importance Ranking (AAAI 2025 · similarity 0.86)
- INQUIRE: A Natural World Text-to-Image Retrieval Benchmark (NeurIPS 2024 · similarity 0.85)
- Semantics Distortion and Style Matter: Towards Source-free UDA for Panoramic Segmentation (CVPR 2024 · similarity 0.85)
- MMIU: Multimodal Multi-image Understanding for Evaluating Large Vision-Language Models (ICLR 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.
