shreshthsaini/brightrate-study-gemma4-12b-multiexposure

Gemma-4-12B BrightVQ multi-exposure study adapter

This PEFT adapter is one result from the BrightRate-LM controlled input and scaling study.

Base model

google/gemma-4-12B-it

Input interface

Eight uniformly sampled frames are each rendered at -2, 0, and +2 stops. The 24 images are passed in temporal-major order.

Training data and recipe

This adapter was trained on the content-separated split 0 of BrightVQ. Training uses two epochs, a three-epoch cosine schedule horizon, learning rate 1e-4, micro-batch 1, gradient accumulation 8, and rank-16 LoRA with alpha 32 and dropout 0.05. MOS targets are interpolated across five quality words.

Training data: BrightVQ.

Metrics

On the 420-video split-0 test set: SROCC 0.6985, PLCC 0.7297, KRCC 0.5086, RMSE 9.1756.

Intended use

This adapter is intended for research on no-reference perceptual quality assessment of user-generated HDR video. Scores are not calibrated for other datasets, display pipelines, or video domains.

Code and input construction are available in BrightRate-LM.

Citation

@article{saini2026brightratelm,
  title   = {BrightRate-LM: Representation-Aware Quality Assessment for User-Generated HDR Video},
  author  = {Saini, Shreshth and Wang, Yilin and Birkbeck, Neil and Adsumilli, Balu and Bovik, Alan C.},
  journal = {Machine Vision and Applications},
  year    = {2026},
  note    = {Submitted}
}

Links

Code and evaluation: github.com/shreshthsaini/BrightRate-LM. Dataset: BrightVQ on Hugging Face. Related papers: Beyond8Bits, CVPR 2026 (arXiv 2603.00938) and CHUG, ICIP 2025 (arXiv 2510.09879).