6.1 技术路线
状态：Agent Verified Summary。NotebookLM 正文可作为叙述草稿；正式结构化入口以本文件、
6.1_taxonomy_from_master.md和paper_master_map.csv为准。
NotebookLM 输入策略
- 29 篇筛选后论文全量输入 NotebookLM，避免技术路线遗漏。
- NotebookLM 输出路线 taxonomy、代表论文、从早期算法到 SOTA 的演进、局限性和趋势。
- Agent 再使用下方 coverage 表检查每篇论文是否被覆盖。
Prompt 文件：
supercode/paper_skim/outputs/stage6/panorama_generation/prompts/6.1_技术路线_notebooklm_prompt.mdAgent Verified Summary
结构化证据层以
paper_master_map.csv、paper_claims.csv、paper_dataset_role_map_v2.csv、read_priority.csv为准；NotebookLM 正文只作为叙述草稿。当前paper_metric_result_verified.csv为空，因此所有 SOTA/首次/最优 claim 都保持待复核状态。
Claim 风险表
| claim_type | rows | status |
|---|---|---|
| novelty_claim | 189 | paper_claim_until_metric_supported |
| sota_claim | 99 | paper_claim_until_metric_supported |
| performance_claim | 38 | paper_claim_until_metric_supported |
| significance_claim | 12 | paper_claim_until_metric_supported |
Dataset Gate 摘要
| sota_eligible_role | rows |
|---|---|
| test_eval | 32 |
| ood_eval | 5 |
Legacy Route Labels
旧
route_label仅用于兼容历史筛选/人工标签；正式 taxonomy 使用下方 task / representation / model_family 三轴。
| 路线 | 论文数 |
|---|---|
| panorama image generation | 4 |
| panoramic image generation | 3 |
| text-to-panorama generation | 2 |
| nfov-to-panorama generation | 2 |
| panorama outpainting | 2 |
| multi-view image generation | 2 |
| text-to-panorama generation analysis | 1 |
| multi-view panorama generation | 1 |
| text-or-image-to-panorama generation | 1 |
| conditional panorama image generation | 1 |
| image-to-panorama generation | 1 |
| unified panorama generation | 1 |
| spherical text-to-image synthesis | 1 |
| top-down-to-panorama generation | 1 |
| panoramic image editing | 1 |
| omnidirectional image generation and editing | 1 |
| panorama-to-panorama translation | 1 |
| panoramic video generation | 1 |
| 3d panorama generation | 1 |
| image-or-video-to-360 lifting | 1 |
结构化 Taxonomy
系统 taxonomy 以
paper_master_map.csv为准，分为 task / representation / model_family 三个轴，避免把任务、表征和模型范式混成单一路线。
Primary Task
| primary_task | 论文数 | 代表论文 |
|---|---|---|
| panorama_generation_general | 9 | Diffusion360: Seamless 360 Degree Panoramic Image Generation based on Diffusion Models; Spherical manifold guided diffusion model for panoramic image generation; Conditional Panoramic Image Generation via Masked Autoregressive Modeling; TanDiT: Tangent-Plane Diffusion Transformer for High-Quality 360 {\deg} Panorama Generation; SphereDiff: Tuning-free 360° Static and Dynamic Panorama Generation via Spherical Latent Representation; JoPano: Unified Panorama Generation via Joint Modeling |
| image_or_nfov_to_panorama | 4 | 360-Degree Panorama Generation from Few Unregistered NFoV Images; Panorama Generation From NFoV Image Done Right; CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation; CamFreeDiff: camera-free image to panorama generation with diffusion model |
| text_to_panorama | 4 | Taming Stable Diffusion for Text to 360° Panorama Image Generation; What Makes for Text to 360-degree Panorama Generation with Stable Diffusion?; DiffPano: Scalable and Consistent Text to Panorama Generation with Spherical Epipolar-Aware Diffusion; Spherical Dense Text-to-Image Synthesis |
| multi_view_panorama | 3 | DiffPano++: Scalable and Consistent Multi-View Panorama Generation with Spherical Epipolar-Aware Diffusion; PanoFree: Tuning-Free Holistic Multi-view Image Generation with Cross-View Self-guidance; MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion |
| panorama_editing_or_translation | 3 | SphereDrag: Spherical Geometry-Aware Panoramic Image Editing; Omni2: Unifying Omnidirectional Image Generation and Editing in an Omni Model; 360PanT: Training-Free Text-Driven 360-Degree Panorama-to-Panorama Translation |
| panorama_outpainting | 2 | PanoDiffusion: 360-degree Panorama Outpainting via Diffusion; Spherical-nested diffusion model for panoramic image outpainting |
| panorama_video_generation | 2 | 360dvd: Controllable panorama video generation with 360-degree video diffusion model; 360Anything: Geometry-Free Lifting of Images and Videos to 360° |
| 3d_or_lifting_to_360 | 1 | DreamCube: 3D Panorama Generation via Multi-plane Synchronization |
| top_down_to_panorama | 1 | Top2Pano: Learning to Generate Indoor Panoramas from Top-Down View |
Primary Representation
| primary_representation | 论文数 | 代表论文 |
|---|---|---|
| erp_direct_or_mixed | 15 | Taming Stable Diffusion for Text to 360° Panorama Image Generation; Diffusion360: Seamless 360 Degree Panoramic Image Generation based on Diffusion Models; 360-Degree Panorama Generation from Few Unregistered NFoV Images; Panorama Generation From NFoV Image Done Right; PanoDiffusion: 360-degree Panorama Outpainting via Diffusion; What Makes for Text to 360-degree Panorama Generation with Stable Diffusion? |
| spherical_latent_or_manifold | 6 | Spherical manifold guided diffusion model for panoramic image generation; Spherical-nested diffusion model for panoramic image outpainting; SphereDiff: Tuning-free 360° Static and Dynamic Panorama Generation via Spherical Latent Representation; Spherical Dense Text-to-Image Synthesis; SphereDiffusion: Spherical Geometry-Aware Distortion Resilient Diffusion Model; SphereDrag: Spherical Geometry-Aware Panoramic Image Editing |
| multi_view_crops | 4 | DiffPano++: Scalable and Consistent Multi-View Panorama Generation with Spherical Epipolar-Aware Diffusion; DiffPano: Scalable and Consistent Text to Panorama Generation with Spherical Epipolar-Aware Diffusion; PanoFree: Tuning-Free Holistic Multi-view Image Generation with Cross-View Self-guidance; MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion |
| cubemap | 2 | CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation; DreamCube: 3D Panorama Generation via Multi-plane Synchronization |
| geometry_free_360_lifting | 1 | 360Anything: Geometry-Free Lifting of Images and Videos to 360° |
| tangent_plane | 1 | TanDiT: Tangent-Plane Diffusion Transformer for High-Quality 360 {\deg} Panorama Generation |
Model Family
| model_family | 论文数 | 代表论文 |
|---|---|---|
| stable_diffusion_or_unet_diffusion | 15 | Taming Stable Diffusion for Text to 360° Panorama Image Generation; Diffusion360: Seamless 360 Degree Panoramic Image Generation based on Diffusion Models; PanoDiffusion: 360-degree Panorama Outpainting via Diffusion; What Makes for Text to 360-degree Panorama Generation with Stable Diffusion?; DiffPano++: Scalable and Consistent Multi-View Panorama Generation with Spherical Epipolar-Aware Diffusion; Spherical manifold guided diffusion model for panoramic image generation |
| unknown_or_mixed | 6 | 360-Degree Panorama Generation from Few Unregistered NFoV Images; Panorama Generation From NFoV Image Done Right; Spherical Dense Text-to-Image Synthesis; Top2Pano: Learning to Generate Indoor Panoramas from Top-Down View; DreamCube: 3D Panorama Generation via Multi-plane Synchronization; 360Anything: Geometry-Free Lifting of Images and Videos to 360° |
| diffusion_transformer | 3 | TanDiT: Tangent-Plane Diffusion Transformer for High-Quality 360 {\deg} Panorama Generation; SphereDrag: Spherical Geometry-Aware Panoramic Image Editing; Omni2: Unifying Omnidirectional Image Generation and Editing in an Omni Model |
| training_free_guidance | 3 | SphereDiff: Tuning-free 360° Static and Dynamic Panorama Generation via Spherical Latent Representation; 360PanT: Training-Free Text-Driven 360-Degree Panorama-to-Panorama Translation; PanoFree: Tuning-Free Holistic Multi-view Image Generation with Cross-View Self-guidance |
| masked_autoregressive | 1 | Conditional Panoramic Image Generation via Masked Autoregressive Modeling |
| unified_omni_model | 1 | JoPano: Unified Panorama Generation via Joint Modeling |
精读优先级入口
下表是 Top 15 excerpt；完整 29 篇精读顺序见
read_priority.md。
| 论文 | score | survey_score | evidence_score | band | task | representation | model_family | figure_status | penalty | why_read |
|---|---|---|---|---|---|---|---|---|---|---|
| CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation | 88 | 68 | 20 | high | image_or_nfov_to_panorama | cubemap | stable_diffusion_or_unet_diffusion | confirmed_pipeline | core paper; code available; confirmed pipeline figure; 68 metric rows need ranking QA; cubemap | |
| Conditional Panoramic Image Generation via Masked Autoregressive Modeling | 75 | 70 | 5 | high | panorama_generation_general | erp_direct_or_mixed | masked_autoregressive | confirmed_pipeline | core paper; code available; confirmed pipeline figure; 64 metric rows extracted but not rankable; masked_autoregressive | |
| JoPano: Unified Panorama Generation via Joint Modeling | 75 | 70 | 5 | high | panorama_generation_general | erp_direct_or_mixed | unified_omni_model | confirmed_pipeline | core paper; code available; confirmed pipeline figure; 118 metric rows extracted but not rankable; unified_omni_model | |
| Taming Stable Diffusion for Text to 360° Panorama Image Generation | 65 | 60 | 5 | high | text_to_panorama | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | confirmed_pipeline | core paper; code available; confirmed pipeline figure; 112 metric rows extracted but not rankable | |
| What Makes for Text to 360-degree Panorama Generation with Stable Diffusion? | 65 | 60 | 5 | high | text_to_panorama | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | confirmed_pipeline | core paper; code available; confirmed pipeline figure; 111 metric rows extracted but not rankable | |
| Spherical manifold guided diffusion model for panoramic image generation | 63 | 58 | 5 | high | panorama_generation_general | spherical_latent_or_manifold | stable_diffusion_or_unet_diffusion | confirmed_pipeline | code_url_not_verified | core paper; confirmed pipeline figure; 100 metric rows extracted but not rankable; spherical_latent_or_manifold |
| SphereDiff: Tuning-free 360° Static and Dynamic Panorama Generation via Spherical Latent Representation | 61 | 56 | 5 | high | panorama_generation_general | spherical_latent_or_manifold | training_free_guidance | partial_method_figure | core paper; code available; figure=partial_method_figure; 13 metric rows extracted but not rankable; spherical_latent_or_manifold | |
| CamFreeDiff: camera-free image to panorama generation with diffusion model | 55 | 50 | 5 | high | image_or_nfov_to_panorama | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | confirmed_pipeline | code_url_not_verified | core paper; confirmed pipeline figure; 47 metric rows extracted but not rankable |
| Spherical-nested diffusion model for panoramic image outpainting | 54 | 38 | 16 | medium | panorama_outpainting | spherical_latent_or_manifold | stable_diffusion_or_unet_diffusion | missing_need_manual_check | code_url_not_verified; figure=missing_need_manual_check | core paper; 16 metric rows need ranking QA; spherical_latent_or_manifold |
| TanDiT: Tangent-Plane Diffusion Transformer for High-Quality 360 {\deg} Panorama Generation | 53 | 48 | 5 | medium | panorama_generation_general | tangent_plane | diffusion_transformer | missing_need_manual_check | code_url_not_verified; figure=missing_need_manual_check | core paper; 252 metric rows extracted but not rankable; diffusion_transformer; tangent_plane |
| Spherical Dense Text-to-Image Synthesis | 53 | 48 | 5 | medium | text_to_panorama | spherical_latent_or_manifold | unknown_or_mixed | missing_need_manual_check | figure=missing_need_manual_check | core paper; code available; 892 metric rows extracted but not rankable; spherical_latent_or_manifold |
| SphereDiffusion: Spherical Geometry-Aware Distortion Resilient Diffusion Model | 53 | 48 | 5 | medium | panorama_generation_general | spherical_latent_or_manifold | stable_diffusion_or_unet_diffusion | missing_need_manual_check | figure=missing_need_manual_check | core paper; code available; 30 metric rows extracted but not rankable; spherical_latent_or_manifold |
| 360-Degree Panorama Generation from Few Unregistered NFoV Images | 45 | 40 | 5 | medium | image_or_nfov_to_panorama | erp_direct_or_mixed | unknown_or_mixed | wrong_or_placeholder | figure=wrong_or_placeholder | core paper; code available; 3 metric rows extracted but not rankable |
| Panorama Generation From NFoV Image Done Right | 45 | 40 | 5 | medium | image_or_nfov_to_panorama | erp_direct_or_mixed | unknown_or_mixed | wrong_or_placeholder | figure=wrong_or_placeholder | core paper; code available; 93 metric rows extracted but not rankable |
| PanoDiffusion: 360-degree Panorama Outpainting via Diffusion | 45 | 40 | 5 | medium | panorama_outpainting | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | missing_need_manual_check | figure=missing_need_manual_check | core paper; code available; 140 metric rows extracted but not rankable |
每篇论文归属检查
| paper_id | 论文 | bucket | primary_task | primary_representation | model_family | figure_status | claim_status |
|---|---|---|---|---|---|---|---|
| a1a69b14748af5b3 | Taming Stable Diffusion for Text to 360° Panorama Image Generation | core | text_to_panorama | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | confirmed_pipeline | paper_claim_until_metric_supported |
| 98bafd1887bf33aa | Diffusion360: Seamless 360 Degree Panoramic Image Generation based on Diffusion Models | core | panorama_generation_general | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 7f4864044e04e5ec | 360-Degree Panorama Generation from Few Unregistered NFoV Images | core | image_or_nfov_to_panorama | erp_direct_or_mixed | unknown_or_mixed | wrong_or_placeholder | paper_claim_until_metric_supported |
| 79ab21dbadb25cf0 | Panorama Generation From NFoV Image Done Right | core | image_or_nfov_to_panorama | erp_direct_or_mixed | unknown_or_mixed | wrong_or_placeholder | paper_claim_until_metric_supported |
| ba47839b8289ce8e | PanoDiffusion: 360-degree Panorama Outpainting via Diffusion | core | panorama_outpainting | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 9753801176957726353 | What Makes for Text to 360-degree Panorama Generation with Stable Diffusion? | core | text_to_panorama | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | confirmed_pipeline | paper_claim_until_metric_supported |
| 1507614108164de8 | DiffPano++: Scalable and Consistent Multi-View Panorama Generation with Spherical Epipolar-Aware Diffusion | core | multi_view_panorama | multi_view_crops | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 9033796522063612996 | Spherical manifold guided diffusion model for panoramic image generation | core | panorama_generation_general | spherical_latent_or_manifold | stable_diffusion_or_unet_diffusion | confirmed_pipeline | paper_claim_until_metric_supported |
| 13521719276910748592 | Spherical-nested diffusion model for panoramic image outpainting | core | panorama_outpainting | spherical_latent_or_manifold | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 983cfc7c8bda0d9e | CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation | core | image_or_nfov_to_panorama | cubemap | stable_diffusion_or_unet_diffusion | confirmed_pipeline | paper_claim_until_metric_supported |
| 95bf3d1227a9198f | Conditional Panoramic Image Generation via Masked Autoregressive Modeling | core | panorama_generation_general | erp_direct_or_mixed | masked_autoregressive | confirmed_pipeline | paper_claim_until_metric_supported |
| 29833d4576d49165 | DiffPano: Scalable and Consistent Text to Panorama Generation with Spherical Epipolar-Aware Diffusion | core | text_to_panorama | multi_view_crops | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 4171918228778611344 | CamFreeDiff: camera-free image to panorama generation with diffusion model | core | image_or_nfov_to_panorama | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | confirmed_pipeline | paper_claim_until_metric_supported |
| 4997777058189649127 | TanDiT: Tangent-Plane Diffusion Transformer for High-Quality 360 {\deg} Panorama Generation | core | panorama_generation_general | tangent_plane | diffusion_transformer | missing_need_manual_check | paper_claim_until_metric_supported |
| c040869e37d6b4ff | SphereDiff: Tuning-free 360° Static and Dynamic Panorama Generation via Spherical Latent Representation | core | panorama_generation_general | spherical_latent_or_manifold | training_free_guidance | partial_method_figure | paper_claim_until_metric_supported |
| 9388373ef8d8e684 | JoPano: Unified Panorama Generation via Joint Modeling | core | panorama_generation_general | erp_direct_or_mixed | unified_omni_model | confirmed_pipeline | paper_claim_until_metric_supported |
| 4733265c143da4c5 | Spherical Dense Text-to-Image Synthesis | core | text_to_panorama | spherical_latent_or_manifold | unknown_or_mixed | missing_need_manual_check | paper_claim_until_metric_supported |
| 7660116cdddca12e | Top2Pano: Learning to Generate Indoor Panoramas from Top-Down View | core | top_down_to_panorama | erp_direct_or_mixed | unknown_or_mixed | wrong_or_placeholder | paper_claim_until_metric_supported |
| 17615279564022297245 | Twindiffusion: Enhancing coherence and efficiency in panoramic image generation with diffusion models | core | panorama_generation_general | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | wrong_or_placeholder | paper_claim_until_metric_supported |
| 7458159420666610471 | Multi-scale diffusion: Enhancing spatial layout in high-resolution panoramic image generation | core | panorama_generation_general | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| ec8de83b9dffe9af | SphereDiffusion: Spherical Geometry-Aware Distortion Resilient Diffusion Model | core | panorama_generation_general | spherical_latent_or_manifold | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 11807705132560466246 | SphereDrag: Spherical Geometry-Aware Panoramic Image Editing | adjacent | panorama_editing_or_translation | spherical_latent_or_manifold | diffusion_transformer | missing_need_manual_check | paper_claim_until_metric_supported |
| 6158533852693537814 | Omni2: Unifying Omnidirectional Image Generation and Editing in an Omni Model | adjacent | panorama_editing_or_translation | erp_direct_or_mixed | diffusion_transformer | confirmed_pipeline | paper_claim_until_metric_supported |
| 1647424063381971462 | 360PanT: Training-Free Text-Driven 360-Degree Panorama-to-Panorama Translation | adjacent | panorama_editing_or_translation | erp_direct_or_mixed | training_free_guidance | confirmed_pipeline | paper_claim_until_metric_supported |
| 5251966204911209787 | 360dvd: Controllable panorama video generation with 360-degree video diffusion model | adjacent | panorama_video_generation | erp_direct_or_mixed | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
| 8a685f26f0cd4183 | DreamCube: 3D Panorama Generation via Multi-plane Synchronization | adjacent | 3d_or_lifting_to_360 | cubemap | unknown_or_mixed | missing_need_manual_check | paper_claim_until_metric_supported |
| d8e4f250c9108fee | PanoFree: Tuning-Free Holistic Multi-view Image Generation with Cross-View Self-guidance | adjacent | multi_view_panorama | multi_view_crops | training_free_guidance | confirmed_pipeline | paper_claim_until_metric_supported |
| 6f20f134666d6e99 | 360Anything: Geometry-Free Lifting of Images and Videos to 360° | adjacent | panorama_video_generation | geometry_free_360_lifting | unknown_or_mixed | missing_need_manual_check | paper_claim_until_metric_supported |
| 295fe81508610dea | MVDiffusion: Enabling Holistic Multi-view Image Generation with Correspondence-Aware Diffusion | adjacent | multi_view_panorama | multi_view_crops | stable_diffusion_or_unet_diffusion | missing_need_manual_check | paper_claim_until_metric_supported |
NotebookLM 回填后校对项
- 每篇论文应有唯一
primary_route，可有secondary_route。 - NotebookLM 的 edge / unclassified 归类不能和 agent
core/adjacentbucket 混为一谈。 - 强措辞如 SOTA、首创、完美等只按论文报告处理，精读前不作为最终事实。
- 建议补“任务 × 表征 × 模型范式”矩阵，减少单一路线分类冲突。
- 所有强 claim 默认状态为
paper_claim_until_metric_supported，进入最终综述前必须由 6.4/6.5 或人工精读验证。