Intro to Graphics 21 - Sampling

แชร์
ฝัง

ความคิดเห็น • 14

  • @pawelstolecki3491
    @pawelstolecki3491 5 หลายเดือนก่อน +1

    Awesome vid! It's summarizing dozens of papers in a single video in a very approachable form.

  • @ImaginaryBlend
    @ImaginaryBlend 2 ปีที่แล้ว +5

    These lectures are pure gold. I hope you will get more attention on YT, you deserve it :) Thank you!

  • @isaacdiaz5793
    @isaacdiaz5793 2 ปีที่แล้ว +4

    Man thank you very much for these tutorials they are so awesome!

  • @obayev
    @obayev 4 หลายเดือนก่อน +1

    Thank you for another great lecture! 🙏🏻☺️

  • @darkarchon89
    @darkarchon89 2 ปีที่แล้ว +3

    For path tracing indirect illumination, are hybrid approaches for selecting N viable? For example, the primary ray gets N = 5, the first secondary rays all get N = 2, and every ray after that gets N = 1. Or is that just needless complexity when you can simply take more samples per pixel?

    • @cem_yuksel
      @cem_yuksel  2 ปีที่แล้ว +4

      In general, it is a better idea to use N=1 and start with a new primary ray every time. There are special cases when skipping the primary ray can be preferable (effectively making N>1 for the primary hit point), but not in general.

  • @bimDe2024
    @bimDe2024 2 ปีที่แล้ว

    54:00 When you mentioned "image space denoiser", does it mean after the "model-view-projection" transform when we get the final scene for displaying on the screen, we need to apply this "denoiser" on that space? Are "image space" and "display space" are synonymous?

    • @cem_yuksel
      @cem_yuksel  2 ปีที่แล้ว +2

      No. You render an image and apply the denoiser to the raster image. Some denoisers use more information than just the pixel color, but they all operate on pixels. That is what image-space means.

  • @NguyenTung-id4ib
    @NguyenTung-id4ib 2 ปีที่แล้ว +1

    when we compute the final radiance for the one pixel, do we average all the traced values by the number of SPP?

    • @cem_yuksel
      @cem_yuksel  2 ปีที่แล้ว +4

      Yes, averaging them (by adding them and then dividing the result with the number of samples) would be the simplest solution. There are other methods that perform weighted averages. If you are interested in them, I'd suggest that you look into "reconstruction filters." Most production renderers would support a number of such filters. For example, here are the ones that V-Ray uses: docs.chaos.com/display/VMAX/Image+Filter

    • @NguyenTung-id4ib
      @NguyenTung-id4ib 2 ปีที่แล้ว

      I think many rendering lectures neglect the post-processing part (i.e., mapping HDR radiance images into 0-1 images). In this regard, I have seen some people who do not average them out but handle this during the tone mapping phase. For me, this is the most confusing part when writing a working renderer.

    • @cem_yuksel
      @cem_yuksel  2 ปีที่แล้ว +1

      @@NguyenTung-id4ib Yes, this can get quite a bit more complicated, since reconstruction filters often use samples of neighboring pixels.

  • @quovadisdorime
    @quovadisdorime 2 ปีที่แล้ว

    Hocam Tükçe altyazı en azından..

  • @majidjalili8784
    @majidjalili8784 ปีที่แล้ว

    The pace of presentation is a bit slow. Mainly because you are looking at some screen behind the camera, and it slows you down.