Panoramic Enhancer (PE): Overview & Applications

• Panoramic Enhancer (PE) is a research motif focused on improving 360° imagery by tackling challenges such as seam discontinuity, distortion, and geometric inconsistencies.
• Various formulations include techniques for street-view panorama generation, distortion correction in video, and physics-informed restoration of compact panoramic optics.
• These methods emphasize global coherence, conditional fidelity, and projection/physics awareness to enhance control, image quality, and reliability across applications.

to=arxiv_search.search 天天中彩票中大奖 天天中彩票足球 պարզ ^^^^^^^^33query33 ^^^^^^^^33max_results33 ^^^^^^^^33sort_by33 ^^^^^^^^33sort_order33 to=arxiv_search.search պարզ diýenjson {"^^^^^^^^33query33 to=arxiv_search.search 天天中彩票网络json {"^^^^^^^^33query33 to=arxiv_search.search njhanijson {"^^^^^^^^33query33 to=arxiv_search.search 大发快三怎么json {"^^^^^^^^33query33 to=arxiv_search.search 天天彩票中大奖json {"^^^^^^^^33query33 ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^°: Panoramic Street-View Generation via Local Scenes Diffusion and Probabilistic Prompting^^^^^^^^33query33descending33 Panoramic Enhancer (PE) is a recurring designation in recent panoramic imaging and generation research for modules or full pipelines that improve the quality, coherence, controllability, or geometric fidelity of ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^° data. In different papers, PE refers to a generative enhancer for stitched street-view panoramas, a distortion-aware module for wide-FoV panoramic video, a computational imaging pipeline for compact annular lenses, or a geometry-aware in-context generation system (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33query33^^^^&&&, &&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33max_results33^^^^&&&, &&&^^^^33sort_by33^^^^&&&). Taken together, these usages suggest that PE is best understood as a task-dependent research motif centered on panorama-specific failure modes—seam discontinuity, latitude-dependent distortion, spatially variant blur, controllability loss, and geometric inconsistency—rather than as a single canonical architecture.

^^^^33relevance33^^^^. Terminology, scope, and recurrent design goals

The term “Panoramic Enhancer” is used heterogeneously. In Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^, it is mapped to a system that both enhances stitched panoramas and generates new ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^° street-view samples for autonomous driving; in QuaDreamer, it is a dual-stream subnetwork for correcting panoramic distortions in quadruped-robot video generation; in ACI/PI^^^^33descending33^^^^RNet and PCIE/PART, it denotes restoration pipelines for compact panoramic optics; and in Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^, it denotes a geometry-aware in-context panoramic generation framework (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33query33^^^^&&&, &&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33max_results33^^^^&&&, &&&^^^^33sort_by33^^^^&&&).

Formulation	Primary role	Core mechanisms
Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^	Enhancement and generation of ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^° street-view panoramas	LSDM and PPM
QuaDreamer PE	Distortion-aware panoramic video enhancement	SSM and FFC
ACI / PI^^^^33descending33^^^^RNet	Restoration of PAL annular panoramas	Wave-based simulation and physics-informed restoration
PCIE / PART	Aberration correction and SR&AC for MPIP	PSF map, PFM, PMAB
Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ PE	Geometry-aware in-context panoramic generation	Parallel RGB–depth pretraining, velocity circular padding, token-level concatenation

Across these formulations, three objectives recur. First, PE commonly targets global coherence, especially at the left–right seam and under equirectangular distortion. Second, it often targets conditional fidelity, meaning that layout, depth, masks, text, or motion controls remain valid after enhancement. Third, several PE systems are explicitly projection-aware or physics-aware, incorporating spherical topology, PSF structure, or panoramic camera geometry directly into the model. This suggests that PE research is defined less by a shared backbone than by a shared insistence that panoramic imagery cannot be treated as an ordinary planar image domain.

^^^^33descending33^^^^. Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^: panoramic enhancement for autonomous driving

Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ is presented as “the first panoramic generation method Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ for autonomous driving” and is explicitly mapped to Panoramic Enhancer (PE) for “improving quality, coherence, and controllability of ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^° street-view panoramas” ^^^^^^^^33(Teng et al., 9 Jul 2025)33 It operates both as an enhancer and a generator: starting from stitched panoramas built from multi-pinhole camera rigs, it reduces seam artifacts, corrects spatial misalignments, and improves visual quality, while also hallucinating coherent, high-quality ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^° panoramas conditioned on BEV map, depth, mask, and text.

Its coherence mechanism is the Local Scenes Diffusion Method (LSDM). Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ uses Latent Diffusion Models where images are encoded to latent space with a VAE encoder PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33^^^^ and decoded by PRESERVED_PLACEHOLDER_^^^^33relevance33^^^^. With PRESERVED_PLACEHOLDER_^^^^33descending33^^^^, forward noising and reverse denoising are written as

PRESERVED_PLACEHOLDER_^^^^33query33^^^^

PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33^^^^

The denoising objective is

PRESERVED_PLACEHOLDER_^^^^33max_results33^^^^

LSDM addresses the fact that stitched panoramas mix coherent subregions PRESERVED_PLACEHOLDER_^^^^33sort_by33^^^^ with aliased seam regions PRESERVED_PLACEHOLDER_^^^^33relevance33^^^^. The panorama is treated as a spatially continuous domain on a sphere, and all ^^^^33descending33^^^^D inputs are circularly shifted by a random angle PRESERVED_PLACEHOLDER_^^^^33sort_order33^^^^: PRESERVED_PLACEHOLDER_^^^^33descending33^^^^ This wrap-around diffusion exposes the model to different seam locations and encourages learning continuity across the PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^ boundary. The underlying equirectangular mapping is

PRESERVED_PLACEHOLDER_^^^^33relevance33relevance33^^^^

which makes continuity at PRESERVED_PLACEHOLDER_^^^^33relevance33descending33^^^^ and PRESERVED_PLACEHOLDER_^^^^33relevance33query33^^^^ structurally central rather than incidental.

Its controllability mechanism is the Probabilistic Prompting Method (PPM). Supported controls include BEV layout maps, semantic or mask maps, depth maps from Depth Anything, and text prompts. For prompts PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^ with features PRESERVED_PLACEHOLDER_^^^^33relevance33max_results33^^^^, ^^^^^^^^33relevance33^^^^^^^^ scores define

PRESERVED_PLACEHOLDER_^^^^33relevance33sort_by33^^^^

The instantiated attention further injects depth and mask priors through PRESERVED_PLACEHOLDER_^^^^33relevance33relevance33^^^^ and PRESERVED_PLACEHOLDER_^^^^33relevance33sort_order33^^^^, with

PRESERVED_PLACEHOLDER_^^^^33relevance33descending33^^^^

This is interpreted in the paper as region-wise probabilistic selection of control signals.

The implementation uses a side-controlling U-Net, a frozen VAE encoder/decoder, a BEV encoder, and a CLIP text encoder. Training uses circular rotations, depth and mask priors for attention gating, and ^^^^33descending33(Teng et al., 9 Jul 2025)33^^^^ diffusion steps; the reported dataset size is ^^^^33descending33sort_order33^^^^,^^^^33relevance33query33(Teng et al., 9 Jul 2025)33^^^^ train images and ^^^^33sort_by33^^^^,^^^^33(Teng et al., 9 Jul 2025)33relevance33descending33^^^^ validation images, with training on two NVIDIA A^^^^33sort_by33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^ GPUs for approximately ^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^ days ^^^^^^^^33(Teng et al., 9 Jul 2025)33

On the nuScenes-^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ validation set, Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ reports BRISQUE PRESERVED_PLACEHOLDER_^^^^33descending33(Teng et al., 9 Jul 2025)33^^^^, PIQE PRESERVED_PLACEHOLDER_^^^^33descending33relevance33^^^^, SSIM PRESERVED_PLACEHOLDER_^^^^33descending33descending33^^^^, FID PRESERVED_PLACEHOLDER_^^^^33descending33query33^^^^, Drivable IoU PRESERVED_PLACEHOLDER_^^^^33descending33(Teng et al., 9 Jul 2025)33^^^^, Mean IoU PRESERVED_PLACEHOLDER_^^^^33descending33max_results33^^^^, and Rank PRESERVED_PLACEHOLDER_^^^^33descending33sort_by33^^^^, outperforming adapted baselines in best no-reference quality and top overall score. When used to augment OneBEV segmentation, Drivable IoU improves from PRESERVED_PLACEHOLDER_^^^^33descending33relevance33^^^^ to PRESERVED_PLACEHOLDER_^^^^33descending33sort_order33^^^^ and Mean IoU from PRESERVED_PLACEHOLDER_^^^^33descending33descending33^^^^ to PRESERVED_PLACEHOLDER_^^^^33query33(Teng et al., 9 Jul 2025)33^^^^, whereas baseline synthetic data did not help ^^^^^^^^33(Teng et al., 9 Jul 2025)33 In this formulation, PE is not merely a renderer; it is a panorama-specific control-and-repair mechanism for replacing stitched artifacts with a learned coherent distribution.

^^^^33query33^^^^. QuaDreamer PE: dual-stream correction for wide-FoV panoramic video

QuaDreamer introduces a different use of the term: a Panoramic Enhancer (PE) designed for controllable panoramic video generation for quadruped robots (&&&^^^^33query33^^^^&&&). Its starting point is the equirectangular projection, where longitude PRESERVED_PLACEHOLDER_^^^^33query33relevance33^^^^ and latitude PRESERVED_PLACEHOLDER_^^^^33query33descending33^^^^ map to image coordinates by

PRESERVED_PLACEHOLDER_^^^^33query33query33^^^^

with sphere coordinates

PRESERVED_PLACEHOLDER_^^^^33query33(Teng et al., 9 Jul 2025)33^^^^

Because the surface area element satisfies PRESERVED_PLACEHOLDER_^^^^33query33max_results33^^^^, apparent horizontal stretching grows like PRESERVED_PLACEHOLDER_^^^^33query33sort_by33^^^^ as PRESERVED_PLACEHOLDER_^^^^33query33relevance33^^^^ increases. The paper associates this with latitude-dependent stretching, seam artifacts at PRESERVED_PLACEHOLDER_^^^^33query33sort_order33^^^^, and wide-FoV geometric inconsistencies under robot jitter.

PE is therefore formulated as a distortion-aware, dual-stream module embedded in the QuaDreamer diffusion backbone. The first stream performs spatial–structure correction through State Space Models. For an encoder feature tensor PRESERVED_PLACEHOLDER_^^^^33query33descending33^^^^, the SSM update is

PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^

SSM blocks are injected before the first downsampling and after the final upsampling. The stated effect is long-range, multi-directional stabilization of horizon and large-scale layout, with explicit emphasis on seam continuity and panoramic unwrapping.

The second stream performs frequency–texture refinement through Fast Fourier Convolution. After three successive downsamplings, intermediate residual blocks apply

PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33relevance33^^^^

with a local/global split of PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33descending33^^^^. The global pathway uses spectral convolution: PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33query33^^^^ and the residual form is described as recovering periodic patterns and sharpening textures while avoiding grid artifacts.

This stage-wise collaboration is central to the paper’s definition of PE: early SSM aligns geometry, middle FFC restores detail, and late SSM suppresses seam drift or ringing that may follow frequency enhancement. Unlike Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^, which emphasizes multi-prompt controllability, QuaDreamer PE is explicitly a distortion-correction subnetwork that complements Vertical Jitter Encoding and the Scene-Object Controller (&&&^^^^33query33^^^^&&&).

Training is end-to-end under the standard diffusion denoising objective

PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^

without auxiliary adversarial or explicit geometry or frequency losses. The reported implementation uses Stable Video Diffusion initialization, a frozen CameraCtrl camera encoder, a single NVIDIA A^^^^33sort_by33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^ ^^^^33(Teng et al., 9 Jul 2025)33sort_order33^^^^G GPU, ^^^^33query33descending33(Teng et al., 9 Jul 2025)33^^^^ epochs, ^^^^33descending33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^k steps, ^^^^33relevance33sort_order33^^^^ hours, batch size PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33max_results33^^^^, learning rate PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33sort_by33^^^^, and DPM-Solver with ^^^^33query33(Teng et al., 9 Jul 2025)33^^^^ sampling steps (&&&^^^^33query33^^^^&&&).

Ablations isolate the contribution of PE. Relative to a baseline with FVD PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33relevance33^^^^, LPIPS PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33sort_order33^^^^, and SSIM PRESERVED_PLACEHOLDER_^^^^33(Teng et al., 9 Jul 2025)33descending33^^^^, adding PE yields FVD PRESERVED_PLACEHOLDER_^^^^33max_results33(Teng et al., 9 Jul 2025)33^^^^, LPIPS PRESERVED_PLACEHOLDER_^^^^33max_results33relevance33^^^^, and SSIM PRESERVED_PLACEHOLDER_^^^^33max_results33descending33^^^^. The combined SOC+PE model achieves LPIPS PRESERVED_PLACEHOLDER_^^^^33max_results33query33^^^^, SSIM PRESERVED_PLACEHOLDER_^^^^33max_results33(Teng et al., 9 Jul 2025)33^^^^, and PTrack PRESERVED_PLACEHOLDER_^^^^33max_results33max_results33^^^^. When OmniTrack is augmented with QuaDreamer data, HOTA improves by PRESERVED_PLACEHOLDER_^^^^33max_results33sort_by33^^^^ and MOTA by PRESERVED_PLACEHOLDER_^^^^33max_results33relevance33^^^^ over “Real Only” (&&&^^^^33query33^^^^&&&). In this branch of the literature, PE denotes a panorama-native corrective prior for wide-FoV video rather than a standalone generator.

^^^^33(Teng et al., 9 Jul 2025)33^^^^. Computational-imaging PE: PAL restoration, PSF priors, and minimalist optics

A restoration-oriented branch of PE is represented by Annular Computational Imaging (ACI) and by the Panoramic Computational Imaging Engine (PCIE). In these papers, the central problem is not promptable generation but recovery of high-quality panoramas from minimalist panoramic optics with severe spatially variant degradation (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33max_results33^^^^&&&).

In ACI, PE is defined as a computational imaging pipeline that restores raw annular panoramas captured by compact Panoramic Annular Lenses (PALs) (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&). The forward model is

PRESERVED_PLACEHOLDER_^^^^33max_results33sort_order33^^^^

with spatially variant PSF PRESERVED_PLACEHOLDER_^^^^33max_results33descending33^^^^. Wave-optics simulation uses an annular pupil, Fresnel diffraction, and Zernike wavefront aberration

PRESERVED_PLACEHOLDER_^^^^33sort_by33(Teng et al., 9 Jul 2025)33^^^^

with PSF obtained as

PRESERVED_PLACEHOLDER_^^^^33sort_by33relevance33^^^^

The simulation samples PRESERVED_PLACEHOLDER_^^^^33sort_by33descending33^^^^ at PRESERVED_PLACEHOLDER_^^^^33sort_by33query33^^^^ and wavelengths PRESERVED_PLACEHOLDER_^^^^33sort_by33(Teng et al., 9 Jul 2025)33^^^^ nm at PRESERVED_PLACEHOLDER_^^^^33sort_by33max_results33^^^^ nm, keeps the first ^^^^33query33relevance33^^^^ Zernike terms, and randomizes coefficients by PRESERVED_PLACEHOLDER_^^^^33sort_by33sort_by33^^^^ to bridge the synthetic-to-real gap (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&).

The restoration network, PI^^^^33descending33^^^^RNet, uses two U-Nets in cascade with a physics-informed single-pass engine, a Physics-informed Bridge, and Dynamic Deformable Kernel Prediction. The total loss is

PRESERVED_PLACEHOLDER_^^^^33sort_by33relevance33^^^^

with PRESERVED_PLACEHOLDER_^^^^33sort_by33sort_order33^^^^. On DIVPano validation, PI^^^^33descending33^^^^RNet reports PSNR PRESERVED_PLACEHOLDER_^^^^33sort_by33descending33^^^^ dB and SSIM PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^, outperforming SRN, NAFNet, HINet, DeepRFT, and KPN; runtime is approximately PRESERVED_PLACEHOLDER_^^^^33relevance33relevance33^^^^ s per PRESERVED_PLACEHOLDER_^^^^33relevance33descending33^^^^ panorama on an RTX ^^^^33query33(Teng et al., 9 Jul 2025)33descending33(Teng et al., 9 Jul 2025)33^^^^ (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&).

PCIE/PART advances a related but distinct formulation for a Minimalist Panoramic Imaging Prototype (MPIP) with fewer than three spherical lenses (&&&^^^^33max_results33^^^^&&&). Here, the image formation model is

PRESERVED_PLACEHOLDER_^^^^33relevance33query33^^^^

with Zernike-based wavefront

PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^

PE is built around a PSF-aware Aberration-image Recovery Transformer (PART). A compressed PSF map is constructed by

PRESERVED_PLACEHOLDER_^^^^33relevance33max_results33^^^^

PSF features then condition both a PSF-aware Feature Modulator (PFM) and a PSF-aware Mix-Attention Block (PMAB). The PFM computes

PRESERVED_PLACEHOLDER_^^^^33relevance33sort_by33^^^^

while PMAB mixes standard window attention and varied-size attention: PRESERVED_PLACEHOLDER_^^^^33relevance33relevance33^^^^

PRESERVED_PLACEHOLDER_^^^^33relevance33sort_order33^^^^

The reported pipelines are AC and SR&AC. On PALHQ-SynMPIP-P^^^^33relevance33^^^^ for AC, PART reports PSNR PRESERVED_PLACEHOLDER_^^^^33relevance33descending33^^^^, SSIM PRESERVED_PLACEHOLDER_^^^^33sort_order33(Teng et al., 9 Jul 2025)33^^^^, LPIPS PRESERVED_PLACEHOLDER_^^^^33sort_order33relevance33^^^^, and FID PRESERVED_PLACEHOLDER_^^^^33sort_order33descending33^^^^; on PALHQ-SynMPIP-P^^^^33descending33^^^^ for AC, PSNR is PRESERVED_PLACEHOLDER_^^^^33sort_order33query33^^^^, SSIM PRESERVED_PLACEHOLDER_^^^^33sort_order33(Teng et al., 9 Jul 2025)33^^^^, LPIPS PRESERVED_PLACEHOLDER_^^^^33sort_order33max_results33^^^^, and FID PRESERVED_PLACEHOLDER_^^^^33sort_order33sort_by33^^^^. On synthetic SR&AC, PART reports PSNR PRESERVED_PLACEHOLDER_^^^^33sort_order33relevance33^^^^, SSIM PRESERVED_PLACEHOLDER_^^^^33sort_order33sort_order33^^^^, LPIPS PRESERVED_PLACEHOLDER_^^^^33sort_order33descending33^^^^, and FID PRESERVED_PLACEHOLDER_^^^^33descending33(Teng et al., 9 Jul 2025)33^^^^. On RealMPIP^^^^33query33^^^^K-AC, OIQE rises to PRESERVED_PLACEHOLDER_^^^^33descending33relevance33^^^^, which the paper reports as the best result (&&&^^^^33max_results33^^^^&&&).

Taken together, ACI/PI^^^^33descending33^^^^RNet and PCIE/PART define a physically grounded understanding of PE: enhancement is treated as inversion of a panoramic optical forward model with explicit priors on PSFs, annular geometry, and sensor behavior.

^^^^33max_results33^^^^. Geometry-aware and in-context PE: the Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ formulation

Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ uses PE in yet another sense: a geometry-aware in-context panoramic generation system built around pretraining and unified fine-tuning (&&&^^^^33sort_by33^^^^&&&). It targets style transfer, inpainting, outpainting, and editing on equirectangular panoramas, with main experiments at PRESERVED_PLACEHOLDER_^^^^33descending33descending33^^^^ and ablations at PRESERVED_PLACEHOLDER_^^^^33descending33query33^^^^.

The pretraining stage uses parallel RGB–depth generation with a Flow Transformer built on FLUX.^^^^33relevance33^^^^-dev. RGB and depth latents are concatenated,

PRESERVED_PLACEHOLDER_^^^^33descending33(Teng et al., 9 Jul 2025)33^^^^

and trained by flow matching on the linear interpolant

PRESERVED_PLACEHOLDER_^^^^33descending33max_results33^^^^

with objective

PRESERVED_PLACEHOLDER_^^^^33descending33sort_by33^^^^

To prevent modality collapse, Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ introduces

PRESERVED_PLACEHOLDER_^^^^33descending33relevance33^^^^

Its seam mechanism is velocity circular padding. Rather than only padding image columns, the method pads both interpolated latents and target velocities: PRESERVED_PLACEHOLDER_^^^^33descending33sort_order33^^^^

PRESERVED_PLACEHOLDER_^^^^33descending33descending33^^^^

The paper argues that this exposes wrap-around adjacency directly in the learned velocity field and improves seam continuity quantitatively and qualitatively. Its best reported LRCE-RGB is PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^ (&&&^^^^33sort_by33^^^^&&&).

Fine-tuning removes the depth branch and uses token-level concatenation: PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33relevance33^^^^ with positional offsets separating target and context roles. This single model supports the four downstream tasks through unified conditioning, rather than training distinct panorama-specific models for each task.

The data scale is unusually large for a panoramic setting. Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^Dataset contains ^^^^33relevance33^^^^M paired panoramic samples, consisting of a ^^^^33relevance33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^K pilot set of RGB–depth panoramas and ^^^^33descending33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^K downstream in-context samples: Outpainting PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33descending33^^^^, Inpainting PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33query33^^^^, Style transfer PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^, and Editing PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33max_results33^^^^. Pseudo-depth is obtained from DAP and clipped at ^^^^33relevance33(Teng et al., 9 Jul 2025)33(Teng et al., 9 Jul 2025)33^^^^ meters for outdoor scenes and ^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^ meters for indoor scenes before normalization (&&&^^^^33sort_by33^^^^&&&).

For text-to-panorama generation, the paper reports FAED PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33sort_by33^^^^ and IS PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33relevance33^^^^ as best, with FID PRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33sort_order33^^^^, FIDPRESERVED_PLACEHOLDER_^^^^33relevance33(Teng et al., 9 Jul 2025)33descending33^^^^, FIDPRESERVED_PLACEHOLDER_^^^^33relevance33relevance33(Teng et al., 9 Jul 2025)33^^^^, QAPRESERVED_PLACEHOLDER_^^^^33relevance33relevance33relevance33^^^^, QAPRESERVED_PLACEHOLDER_^^^^33relevance33relevance33descending33^^^^, BRISQUE PRESERVED_PLACEHOLDER_^^^^33relevance33relevance33query33^^^^, and NIQE PRESERVED_PLACEHOLDER_^^^^33relevance33relevance33(Teng et al., 9 Jul 2025)33^^^^. In a user study with ^^^^33relevance33relevance33^^^^ participants and ^^^^33relevance33(Teng et al., 9 Jul 2025)33^^^^ images, the method leads in boundary continuity, panorama awareness, and overall quality (&&&^^^^33sort_by33^^^^&&&). In this usage, PE denotes a panorama-specific generative prior that is geometry-aware before it is task-aware.

^^^^33sort_by33^^^^. Evaluation regimes, applications, and open issues

PE systems are evaluated along several non-interchangeable axes. Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ combines image quality assessment, controllability, and downstream Bird’s Eye View segmentation through SSIM, FID, BRISQUE, PIQE, Drivable IoU, and Mean IoU ^^^^^^^^33(Teng et al., 9 Jul 2025)33 QuaDreamer measures video fidelity and control through FVD, LPIPS, SSIM, PTrack, HOTA, and MOTA (&&&^^^^33query33^^^^&&&). ACI/PI^^^^33descending33^^^^RNet and PCIE/PART evaluate restoration using PSNR, SSIM, LPIPS, FID, OIQE, BRISQUE, and NIQE (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33max_results33^^^^&&&). Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ adds panorama-aware metrics such as FAED and LRCE-RGB, which are explicitly designed to capture ERP-specific artifacts that perspective-trained metrics may underweight (&&&^^^^33sort_by33^^^^&&&).

The application domains are correspondingly broad. Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ is tied to autonomous driving data regeneration and BEV segmentation ^^^^^^^^33(Teng et al., 9 Jul 2025)33 QuaDreamer is aimed at quadruped robots and ^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^° multi-object tracking under vertical jitter (&&&^^^^33query33^^^^&&&). ACI/PI^^^^33descending33^^^^RNet and PCIE/PART target mobile and wearable panoramic imaging with PAL or MPIP hardware (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33max_results33^^^^&&&). Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ targets in-context generation tasks, including style transfer, inpainting, outpainting, and editing (&&&^^^^33sort_by33^^^^&&&). This distribution of tasks indicates that PE has become a bridge term linking restoration, generation, and control in panoramic research.

Several limitations recur across formulations. Percep^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ notes domain gaps between stitched panoramas and real panoramic cameras, hallucination risks in occluded or aliased regions, moving-object artifacts, calibration sensitivity, and controllability limits for precise road topology; it also states that generated data should be flagged as synthetic and that perception systems should not be trained solely on hallucinated scenes without adequate validation ^^^^^^^^33(Teng et al., 9 Jul 2025)33 QuaDreamer notes weaker performance at panoramic boundaries, difficulty under extreme jitter and blur, near-field parallax, and sensitivity to equirectangular format and seam alignment (&&&^^^^33query33^^^^&&&). ACI and PCIE both identify synthetic-to-real gaps, sensitivity to ISP mismatch or calibration error, and residual failures under extreme aberration, low light, or unmodeled noise (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33max_results33^^^^&&&). Canvas^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ identifies underrepresented categories such as high-resolution faces and dense text signage, especially in high-distortion ERP regions (&&&^^^^33sort_by33^^^^&&&).

Related research reinforces these themes. PanoWorld reframes panoramic video generation as geometry- and dynamics-consistent latent state modeling with depth and trajectory consistency losses, emphasizing that panoramic generation should be treated as a geometric modeling problem (&&&^^^^33(Teng et al., 9 Jul 2025)33max_results33^^^^&&&). SphereDrag addresses boundary discontinuity, trajectory deformation, and uneven pixel density in panoramic editing through Adaptive Reprojection, Great-Circle Trajectory Adjustment, and Spherical Search Region Tracking (&&&^^^^33(Teng et al., 9 Jul 2025)33sort_by33^^^^&&&). Pano^^^^33query33sort_by33(Teng et al., 9 Jul 2025)33^^^^ extends stitching into ^^^^33query33^^^^D photogrammetric space for globally consistent alignment across many views (&&&^^^^33(Teng et al., 9 Jul 2025)33relevance33^^^^&&&). These adjacent directions do not use PE identically, but they converge on the same conclusion: panoramic enhancement is fundamentally conditioned by spherical geometry, projection nonuniformity, and cross-view consistency.

Future directions stated across the literature include better transfer of pinhole annotations to panoramic domains, stronger prompt utilization and multi-view geometry constraints, extension to panoramic videos with temporal coherence, and richer geometry-aware pretraining (&&&^^^^33(Teng et al., 9 Jul 2025)33^^^^&&&, &&&^^^^33sort_by33^^^^&&&, &&&^^^^33(Teng et al., 9 Jul 2025)33max_results33^^^^&&&). A plausible implication is that the term “Panoramic Enhancer” will continue to denote systems that couple panorama-specific inductive bias with either generative control, optical inversion, or geometric regularization, rather than converging on a single universal model class.