Colour map is a bit different from TA solution, but I hope TAs wont mind
Center 0.25,0.25,0.00
Side 2.00,1.50,1.50
Without view dependence
Left to right : Without view dependence, With view dependence
As you can see the rendering on the board in the front is a bit inconsistent when view dependence is not being used
If we increase our dependence on view direction to generate per segment density and colour, that could lead to overfitting of model. By increasing dependence if you mean passing viewing direction in the first layer along with position as opposed to original nerf where its passed in the final few layers, the model can theoretically learn small images patches placed in front of the camera locations to overfit on the artifacts.