We present Intrinsics in Flux (InFlux) and InFlux++, which together form a comprehensive data suite for dynamic intrinsics prediction in videos whose camera intrinsics change over time as the camera zooms or refocuses. Our data suite has two parts: a large-scale real-world benchmark for evaluation and a procedurally generated synthetic training set.
We present InFlux++, consisting of two components. InFlux++ Real is a large-scale real-world benchmark for evaluating dynamic intrinsics prediction, with 514K+ frames captured across 334 high-resolution videos. It provides per-frame ground truth intrinsics for videos with changing zoom and focus. Compared to InFlux, InFlux++ spans a wider range of scenes and includes longer video sequences with substantially more camera translation and parallax.
InFlux++ Synth is a large-scale procedurally generated synthetic video dataset with 441K+ annotated frames from 1,841 high-resolution videos, providing accurate per-frame ground truth intrinsics for training dynamic intrinsics prediction models; a subset additionally includes per-frame camera pose, depth, and surface normals. The videos feature rich intrinsics diversity by changing both camera zoom and focus over time, along with dynamic objects and realistic rendering effects such as camera distortion, lens breathing, and defocus blur.
When evaluating existing intrinsics prediction methods on InFlux and InFlux++ Real, we find that they continue to struggle on real videos with dynamic intrinsics. But finetuning existing intrinsics prediction methods on InFlux++ Synth consistently improves focal length estimation across both InFlux++ Real and InFlux, suggesting that synthetic supervision is promising for RGB-based intrinsics prediction.
Erich Liang,
Caleb Kha-Uong*,
Chinmaya Saran*,
Sreemanti Dey*,
David W. Liu,
Junhan Ouyang,
Benjamin Zhou,
Jia Deng
* Equal contribution
European Conference on Computer Vision (ECCV), 2026
We present InFlux, a real-world benchmark that provides per-frame ground truth intrinsics annotations for videos with dynamic intrinsics. Compared to prior benchmarks, InFlux captures a wider range of intrinsic variations and scene diversity, featuring 143K+ annotated frames from 386 high-resolution indoor and outdoor videos with dynamic camera intrinsics. To ensure accurate per-frame intrinsics, we build a comprehensive lookup table of calibration experiments and extend the Kalibr toolbox to improve its accuracy and robustness. Using our benchmark, we evaluate existing baseline methods for predicting camera intrinsics and find that most struggle to achieve accurate predictions on videos with dynamic intrinsics.
Erich Liang,
Roma Bhattacharjee*,
Sreemanti Dey*,
Rafael Moschopoulos,
Caitlin Wang,
Michel Liao,
Grace Tan,
Andrew Wang,
Karhan Kayan,
Stamatis Alexandropoulos,
Jia Deng
* Equal contribution
Neural Information Processing Systems Datasets and Benchmarks Track (NeurIPS), 2025
If you use our benchmark, data, evaluation server, or methods in your work, please cite the corresponding paper or papers.
@misc{liang2026influxrealsyntheticdata,
title={InFlux++: Real and Synthetic Data for Estimating Dynamic Camera Intrinsics},
author={Erich Liang and Caleb Kha-Uong and Chinmaya Saran and Sreemanti Dey and David W. Liu and Junhan Ouyang and Benjamin Zhou and Jia Deng},
year={2026},
eprint={2607.05389},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2607.05389},
}
@inproceedings{liang2025influx,
author = {Liang, Erich and Bhattacharjee, Roma and Dey, Sreemanti and Moschopoulos, Rafael and Wang, Caitlin and Liao, Michel and Tan, Grace and Wang, Andrew and Kayan, Karhan and Alexandropoulos, Stamatis and Deng, Jia},
booktitle = {Advances in Neural Information Processing Systems},
editor = {D. Belgrave and C. Zhang and H. Lin and R. Pascanu and P. Koniusz and M. Ghassemi and N. Chen},
pages = {},
publisher = {Curran Associates, Inc.},
title = {InFlux: A Benchmark for Self-Calibration of Dynamic Intrinsics of Video Cameras},
url = {https://proceedings.neurips.cc/paper_files/paper/2025/file/8a8eca190088852067b4e8cc1b907122-Paper-Datasets_and_Benchmarks_Track.pdf},
volume = {38},
year = {2025}
}
This work was partially supported by the National Science Foundation. We thank our friends and colleagues at Princeton University for their help filming the real-world benchmarks.