ymhe12
commented
1 month ago Hi, thank you for your interest in our work!
Is the Blinn-Phong model more suitable for representing the shadow than the other models?
In this aspect, they are essentially the same. Both the Blinn-Phong model and other models, like the rendering function, require occlusion information (for example, the occlusion term in GS-IR) for training. This occlusion term is trained alongside the parameters of the lighting model. We chose the Blinn-Phong model for its simplicity and explicitness, which enhances its generalization ability, especially in out-of-distribution (OOD) scenarios.
In algorithm 1, you mentioned about the shadow coefficient phi. What is this coefficient called in other equations, since I cannot find phi in the rest of the paper.
Sorry for the confusion. The shadow coefficient $\phi$ is a factor applied to the light intensity $T_i^\text{light}$.
Similar to the idea of camera-to-point accumulated transmittance in 3DGS, our light intensity defines the accumulated transmittance of light intensity along the light-to-point ray, thus is influenced by the magnitude of the opacities. To facilitate learning and convergence, we added a coefficient $\phi$ to this term, referred to as the shadow coefficient. Thus the shadow computing equation (Eq.4) should be updated as:
$$ L_p = k_aI_a + \phi Ti^\text{light} \times \sum\limits{m \in \text{lights}} (k_dI_d + k_sI_s). $$
Let me know if you have other questions.
KhanhNgoDuy
Hi, I have a few questions about the paper.
phi. What is this coefficient called in other equations, since I cannot findphiin the rest of the paper.Thank you in advance, I really appreciate your answer.