Non skip pdf in lights HittableList

[niccolot]

Hello, if i add to the lights list an object that is not a metal or dielectric the image produced is like

i' ve obtained this by putting

// in the original main is glass
auto nonglass = make_shared<lambertian>(color(0.8, 0.2, 0.2));
world.add(make_shared<sphere>(point3(190,90,190), 90, nonglass));

// Light Sources
hittable_list lights;
auto m = shared_ptr<material>();
lights.add(make_shared<quad>(point3(343,554,332), vec3(-130,0,0), vec3(0,0,-105), m));
lights.add(make_shared<sphere>(point3(190, 90, 190), 90, m));

in the main.cpp of therestofyourlife part.

I may not have understood correctly this part of the code and thus it may not be an actual issue but still i wanted to ask this since also a material like a lambertian is still emitting some reflected light

[niccolot]

I' ve looked into it a bit and found out that what's causing the problem is the sphere part of the HittablePDF list belonging to lights:

// main.cpp
HittableList lights;
lights.add(std::make_shared<Quad>(Vec3(213, 554, 227), Vec3(130, 0, 0), Vec3(0, 0, 105), empty_mat));
lights.add(std::make_shared<Sphere>(Vec3(190, 90, 190), 90, empty_mat));

if i select just the sphere part

// camera.cpp
Color color_from_scatter = Color();
    for (const auto& ray_t : srec.scattered_rays) {
        if (ray_t.skip_pdf || ray_t.pdf == nullptr) {
            color_from_scatter += ray_t.attenuation * ray_color(ray_t.skip_pdf_ray, depth-1, world, lights);
        } else {
            auto light_ptr = std::make_shared<HittablePDF>(lights, rec.point()); // lights is the problematic part
            MixturePDF p(light_ptr, ray_t.pdf); 
            Ray scattered = Ray(rec.point(), p.generate(), r.time());
            double pdf_value = p.value(scattered.direction(), r.direction(), -w);
            double scattering_pdf = rec.material()->scattering_pdf(r, rec, scattered, -w);
            Color sample_color = ray_color(scattered, depth-1, world, lights);
                
            color_from_scatter += (ray_t.attenuation * scattering_pdf * sample_color) / pdf_value;
        }
    }

// hittable.pdf
double HittableList::pdf_value(const Vec3& origin, const Vec3& direction) const {
    auto weight = 1.0 / objects.size();
    double sum{};

    for (const auto& obj : objects) {
        sum += weight * obj->pdf_value(origin, direction);
    }

    //return sum;
    return objects[1]->pdf_value(origin, direction);
}

Vec3 HittableList::random(const Vec3& origin) const {
    auto int_size = int(objects.size());
    //return objects[random_int(0, int_size-1)]->random(origin);
return objects[1]->random(origin);
}

i get

while e.g. with only the light part (index 0) i get a decent image

by comparison, excluding the sphere from the lights list i get

By checking some values between random and pdf_value i saw that sometimes the vectors are NaNs:

objects[1]->random(origin): -nan -nan -nan

but adding some temporary control like

Vec3 HittableList::random(const Vec3& origin) const {
    auto int_size = int(objects.size());
    auto vec = objects[random_int(0, int_size-1)]->random(origin);

    while (vec.is_nan()) {
        vec = objects[random_int(0, int_size-1)]->random(origin);
    }

    return vec;
}

// vector.h
bool is_nan() const {
            return std::isnan(e[0]) || std::isnan(e[1]) || std::isnan(e[2]);
}

had no change, so this does not seem to be the problem.

[hollasch]

Thank you for grinding on this!

[niccolot]

Hello, i went back to the problem and managed to solve it by adding a control to the sampling of the sphere and the previously mentioned for the random method of hittable_list:

double pdf_value(const point3& origin, const vec3& direction) const override {
        // This method only works for stationary spheres.

        hit_record rec;
        if (!this->hit(ray(origin, direction), interval(0.001, infinity), rec))
            return 0;

        auto cos_theta_max = sqrt(1 - radius*radius/(center.at(0) - origin).length_squared());
        auto solid_angle = 2*pi*(1-cos_theta_max);

        if (std::isnan(solid_angle) || std::fabs(solid_angle) < 1e-3) {
            return 0;
        } else {
            return  1 / solid_angle;
        }
    }

Now the image does not turn black when a "pdf type" material (e.g. lambertians) are added to the lights list.

I'm including some images that shows the cornell box with and without the sphere in the lights list:

without

with

As you can see the sphere is better highlighted, although there are some black pixels, a thing that i actually don't see in my own implementation, so there might be something else behind it

image from my repo

I' ve opened a draft pull request #1674 in case you are interested, it would be great to contribute to the project even a little bit!

In case tell me if i should close this issue.