Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Conditional Panoramic Image Generation via Masked Autoregressive Modeling
Authors: Chaoyang Wang, Xiangtai Li, Lu Qi, Xiaofan Lin, Jinbin Bai, Qianyu Zhou, Yunhai Tong
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate competitive performance in text-to-image generation and panorama outpainting tasks while showcasing promising scalability and generalization capabilities. ... 4 Experiment |
| Researcher Affiliation | Collaboration | 1School of Intelligence Science and Technology, Peking University 2Insta360 Research 3National University of Singapore 4The University of Tokyo |
| Pseudocode | No | The paper describes the methodology using text and diagrams (Figure 2), but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Project Page: https://wang-chaoyang.github.io/project/par. and in the Neur IPS Paper Checklist, question 5: "Does the paper provide open access to the data and code..." Answer: [Yes] Justification: The code is released. |
| Open Datasets | Yes | We mainly use Matterport3D [6] for comparisons. ... SUN360 [66] is used for evaluation... We include additional experiments using the Structured3D [78] dataset... |
| Dataset Splits | Yes | The split of the training and validation set follows Pan Fusion [74]. We sampled 9000 images for training and 1000 for testing. |
| Hardware Specification | Yes | It takes about 2 days to train the 1.4B model on 8 NVIDIA A100 GPUs. ... using one NVIDIA A100 GPU with a batch size of 8. |
| Software Dependencies | No | We employ an Adam W optimizer [35]. We utilize NOVA [14] as the initialization... The paper does not provide specific version numbers for software dependencies such as programming languages or deep learning frameworks. |
| Experiment Setup | Yes | We utilize NOVA [14] as the initialization and set the resolution as 512 1024. Our model is trained for 20K iterations with a batch size 32. We employ an Adam W optimizer [35]. The learning rate is 5 10 5 with the linear scheduler. In the inference stage, we set the CFG [23] coefficient as 5. ... The sampling step for PAR is set to 64 and r = 0.125 by default. ... The denoising steps for MLP are 25. ... λ is set as 0.1 in our experiments. |