Can we implement supersampling, so that along the border of the shadow we have extra rays generated, and far away we don't bother doing a full 100?
I think the professor mentioned that this image is already generated by supersampling, with multiple rays in different directions cast from each ground pixel. That contributes to why the image is so noisy, since many of those random rays point off into nothing in space.
Your idea about taking more samples along borders is really interesting though, since it would help increase the quality of the picture where it matters most. What do you think would be a good way to figure out where the borders are?
I think we can first do point sample and get a clear edge of the shadow. Then the algorithm identifies pixels near the boarder. For those pixels, we again perform Monte Carlo Sampling.