Cascading reflections in dielectric, differs from book render but code same

[ayaanmaan]

Hello. I am getting this kind of effect in the renders compared to what is shown in the book:


and similarly


Also notice a distinctive straight line in the blue sphere in first image and brown metal sphere in second (mine).

Does anybody have an explanation for why this might be happening?
My code for reference: https://github.com/ayaanmaan/week-tracing

[ayaanmaan]

I found the problem! It was a line in camera.h:
if (world.hit(r, interval(0, infinity), rec)) {

I compared with the original repo and found the difference here:
if (world.hit(r, interval(0.001, infinity), rec)) {

Why does this lead to the weird behaviour though, I still don't know? Darkened colors, weird artifacts, cascading images, longer render times (almost double at low pixel sample sizes).

For another example, observe the final renders

my code (before fix) took around 3 hours to render at samples_per_pixel = 1000 and max_depth = 1000 (╥﹏╥)

fast render at samples_per_pixel = 10 and max_depth = 20 from original repo's code

[armansito]

The ray origin is never precisely "on the surface" of the sphere due to a lack of high floating point precision. It is possible for an intersection point to sometimes be slightly "below" or "above" the surface. The 0.001 adds a small margin along the direction such that intersections that are very close to the ray origin get rejected.

Otherwise, it is possible for the new intersection point to end up on the same surface patch as the originating point.

For a reflection ray, imagine re-intersecting the same sphere from slightly below the surface. Similarly for a refraction, you can end up hitting the same surface from the same side multiple times.