Bias for Action: Video Implicit Neural Representations with Bias Modulation

Alper Kayabasi1, Anil Kumar Vadathya2, Guha Balakrishnan3, Vishwanath Saragadam1
1University of California Riverside
2Neal Cancer Center,Houston Methodist Hospital
3Rice University
Paper arXiv

We introduce a novel INR architecture for video modeling. Fundamentally, an INR can be interpreted as a basis function expansion of a signal, where weights determine the shape of the basis functions, while biases control their translation. Building on this, our design shares weights across the video to maintain consistent basis functions while assigning a unique bias to each frame. This approach effectively models the shift of basis functions, capturing motion within the video.

Dramatic Interpolation Task x10
Video Denoising
Video Inpainting

Abstract

We propose a new continuous video modeling framework based on implicit neural representations (INRs) called ActINR. At the core of our approach is the observation that INRs can be considered as a learnable dictionary, with the shapes of the basis functions governed by the weights of the INR, and their locations governed by the biases. Given compact non-linear activation functions, we hypothesize that an INR's biases are suitable to capture motion across images, and facilitate compact representations for video sequences. Using these observations, we design ActINR to share INR weights across frames of a video sequence, while using unique biases for each frame. We further model the biases as the output of a separate INR conditioned on time index to promote smoothness. By training the video INR and this bias INR together, we demonstrate unique capabilities, including 10x video slow motion, 4x spatial super resolution along with 2x slow motion, denoising, and video inpainting. ActINR performs remarkably well across numerous video processing tasks (often achieving more than 6dB improvement), setting a new standard for continuous modeling of videos.

Qualitative Results and Comparisons

Qualitative results for 2× interpolation


Qualitative Results for 10× interpolation.


BibTeX

@misc{kayabasi2025bias,
      title={Bias for Action: Video Implicit Neural Representations with Bias Modulation}, 
      author={Alper Kayabasi, Anil Kumar Vadathya, Guha Balakrishnan, Vishwanath Saragadam},
      year={2025},
      eprint={2501.09277},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}