introvoyz041/LowLevelEval

Is Nano Banana Pro a Low-Level Vision All-Rounder? A Comprehensive Evaluation on 14 Tasks and 40 Datasets

Paper | Project page | GitHub Repo

This repository hosts the official datasets and inferred results from the technical report: "Is Nano Banana Pro a Low-Level Vision All-Rounder? A Comprehensive Evaluation on 14 Tasks and 40 Datasets."

While commercial text-to-image (T2I) models like Nano Banana Pro excel in creative synthesis, their potential as generalist solvers for traditional low-level vision challenges remains largely underexplored. In this study, we investigate the critical question: Is Nano Banana Pro a Low-Level Vision All-Rounder? We conducted a comprehensive zero-shot evaluation across 14 distinct low-level tasks spanning 40 diverse datasets.

Key Highlights

• Massive Benchmark: Evaluated on 14 low-level vision tasks and 40 datasets.
• Zero-Shot Setting: Utilized simple textual prompts without any fine-tuning.
• The Dichotomy Discovery: We reveal a distinct performance dichotomy:
  • ✅ Superior Subjective Quality: Often hallucinates plausible high-frequency details that surpass specialist models.
  • ❌ Lower Reference-Based Metrics: Lags behind in PSNR/SSIM due to the inherent stochasticity of generative models.

Our extensive analysis identifies Nano Banana Pro as a capable zero-shot contender for low-level vision tasks. While it struggles to maintain the strict pixel-level consistency required by conventional metrics (PSNR/SSIM), it offers superior visual quality, suggesting a need for new perception-aligned evaluation paradigms.

This HuggingFace repository contains the datasets used in this evaluation project along with the corresponding inference results obtained from the Nano Banana Pro.

Citation

If you find this work helpful for your research, please consider citing:

@misc{zuo2025nanobananaprolowlevel,
      title={Is Nano Banana Pro a Low-Level Vision All-Rounder? A Comprehensive Evaluation on 14 Tasks and 40 Datasets}, 
      author={Jialong Zuo and Haoyou Deng and Hanyu Zhou and Jiaxin Zhu and Yicheng Zhang and Yiwei Zhang and Yongxin Yan and Kaixing Huang and Weisen Chen and Yongtai Deng and Rui Jin and Nong Sang and Changxin Gao},
      year={2025},
      eprint={2512.15110},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2512.15110}, 
}