NeRF Director: Revisiting View Selection in Neural Volume Rendering
This addresses a key bottleneck in neural rendering for 3D computer vision, offering an incremental improvement in view selection efficiency.
The paper tackles the problem of selecting training views for neural volume rendering, showing that a simple rotation of test views can change performance rankings of state-of-the-art methods, and proposes a training-free approach based on uniform view coverage that achieves high-quality renderings faster with fewer views.
Neural Rendering representations have significantly contributed to the field of 3D computer vision. Given their potential, considerable efforts have been invested to improve their performance. Nonetheless, the essential question of selecting training views is yet to be thoroughly investigated. This key aspect plays a vital role in achieving high-quality results and aligns with the well-known tenet of deep learning: "garbage in, garbage out". In this paper, we first illustrate the importance of view selection by demonstrating how a simple rotation of the test views within the most pervasive NeRF dataset can lead to consequential shifts in the performance rankings of state-of-the-art techniques. To address this challenge, we introduce a unified framework for view selection methods and devise a thorough benchmark to assess its impact. Significant improvements can be achieved without leveraging error or uncertainty estimation but focusing on uniform view coverage of the reconstructed object, resulting in a training-free approach. Using this technique, we show that high-quality renderings can be achieved faster by using fewer views. We conduct extensive experiments on both synthetic datasets and realistic data to demonstrate the effectiveness of our proposed method compared with random, conventional error-based, and uncertainty-guided view selection.