High-Frequency First: A Two-Stage Approach for Improving Image INR
This work addresses a key limitation in INR methods for image processing, offering a new training-based approach to enhance detail capture, though it appears incremental as it complements existing methods rather than introducing a paradigm shift.
The paper tackles the spectral bias problem in Implicit Neural Representations (INRs) for images, where networks struggle to capture high-frequency details, by proposing a two-stage training strategy that adaptively weights pixels with strong local variations to focus on fine details early, resulting in improved reconstruction quality.
Implicit Neural Representations (INRs) have emerged as a powerful alternative to traditional pixel-based formats by modeling images as continuous functions over spatial coordinates. A key challenge, however, lies in the spectral bias of neural networks, which tend to favor low-frequency components while struggling to capture high-frequency (HF) details such as sharp edges and fine textures. While prior approaches have addressed this limitation through architectural modifications or specialized activation functions, we propose an orthogonal direction by directly guiding the training process. Specifically, we introduce a two-stage training strategy where a neighbor-aware soft mask adaptively assigns higher weights to pixels with strong local variations, encouraging early focus on fine details. The model then transitions to full-image training. Experimental results show that our approach consistently improves reconstruction quality and complements existing INR methods. As a pioneering attempt to assign frequency-aware importance to pixels in image INR, our work offers a new avenue for mitigating the spectral bias problem.