Implementing Frustum Culling in Python Ray Tracing Camera

Question

I'm working on a ray tracing project in Python and have encountered a performance bottleneck. I believe implementing frustum culling in my Camera class could significantly improve rendering times.

Camera:

class Camera():
    def __init__(self, look_from, look_at, screen_width = 400 ,screen_height = 300,  field_of_view = 90., aperture = 0., focal_distance = 1.):
        self.screen_width = screen_width
        self.screen_height = screen_height
        self.aspect_ratio = float(screen_width) / screen_height
        self.look_from = look_from
        self.look_at = look_at
        self.camera_width = np.tan(field_of_view * np.pi/180   /2.)*2.
        self.camera_height = self.camera_width/self.aspect_ratio
        self.cameraFwd = (look_at - look_from).normalize()
        self.cameraRight = (self.cameraFwd.cross(vec3(0.,1.,0.))).normalize()
        self.cameraUp = self.cameraRight.cross(self.cameraFwd)
        self.lens_radius =  aperture / 2.
        self.focal_distance = focal_distance
        self.x = np.linspace(-self.camera_width/2., self.camera_width/2., self.screen_width)
        self.y = np.linspace(self.camera_height/2., -self.camera_height/2., self.screen_height)
        xx,yy = np.meshgrid(self.x,self.y)
        self.x = xx.flatten()
        self.y = yy.flatten()
    def get_ray(self,n):
        x = self.x + (np.random.rand(len(self.x )) - 0.5)*self.camera_width  /(self.screen_width)
        y = self.y + (np.random.rand(len(self.y )) - 0.5)*self.camera_height /(self.screen_height)
        r = np.sqrt(np.random.rand(x.shape[0]))
        phi = np.random.rand(x.shape[0])*2*np.pi
        ray_origin = self.look_from  +   self.cameraRight *r * np.cos(phi)* self.lens_radius   +   self.cameraUp *r * np.sin(phi)* self.lens_radius
        return Ray(origin=ray_origin, dir=(self.look_from  +   self.cameraUp*y*self.focal_distance  +  self.cameraRight*x*self.focal_distance  + self.cameraFwd*self.focal_distance  - ray_origin  ).normalize(), depth=0,  n=n, reflections = 0, transmissions = 0, diffuse_reflections = 0)

Frustum culling is not a suitable approach to increase your ray tracing performance. If you ignore objects outside the view frustum, you will degrade the indirect lighting. In a dynamic scene, this will manifest itself in abrupt changes when new objects enter the frustum. — Stefan Scheller
– Stefan Scheller, Commented Jan 16, 2024 at 15:15

user23257844 · Accepted Answer · 2024-01-17 11:10:39Z

You just need to modify the Camera class as following:

class Camera():
    def __init__(self, look_from, look_at, screen_width = 400 ,screen_height = 300,  field_of_view = 90., aperture = 0., focal_distance = 1.):
        self.screen_width = screen_width
        self.screen_height = screen_height
        self.aspect_ratio = float(screen_width) / screen_height
        self.look_from = look_from
        self.look_at = look_at
        self.camera_width = np.tan(field_of_view * np.pi / 180 / 2.) * 2.
        self.camera_height = self.camera_width / self.aspect_ratio
        self.cameraFwd = (look_at - look_from).normalize()
        self.cameraRight = (self.cameraFwd.cross(vec3(0., 1., 0.))).normalize()
        self.cameraUp = self.cameraRight.cross(self.cameraFwd)
        self.lens_radius = aperture / 2.
        self.focal_distance = focal_distance
        self.near = .1
        self.far = 100.
        self.x = np.linspace(-self.camera_width / 2., self.camera_width / 2., self.screen_width)
        self.y = np.linspace(self.camera_height / 2., -self.camera_height / 2., self.screen_height)
        xx, yy = np.meshgrid(self.x, self.y)
        self.x = xx.flatten()
        self.y = yy.flatten()
    def get_ray(self,n):
        x = self.x + (np.random.rand(len(self.x)) - 0.5) * self.camera_width / (self.screen_width)
        y = self.y + (np.random.rand(len(self.y)) - 0.5) * self.camera_height / (self.screen_height)
        ray_origin = self.look_from + self.cameraRight * x * self.near + self.cameraUp * y * self.near
        return Ray(origin=ray_origin, dir=(self.look_from + self.cameraUp * y * self.focal_distance +self.cameraRight * x * self.focal_distance +self.cameraFwd * self.focal_distance - ray_origin).normalize(), depth=0, n=n, reflections=0, transmissions=0, diffuse_reflections=0)

Collectives™ on Stack Overflow

Implementing Frustum Culling in Python Ray Tracing Camera

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related