Refractive Errrors when Camera is far from (0,0,0) #103005

New Issue

Jack Toth · 2022-12-08T03:32:20+01:00

Jack Toth commented

2022-12-08 03:32:20 +01:00

System Information
Operating system: PC
Graphics card: RTX3080Ti

Blender Version
Broken: 3.31, others
Worked: none

Hi, I am recreating the geometry of optical camera lenses to shoot through for photographic effect. I have noticed that the results look great with my current method when the camera is at location (0,0,0) however when moving to greater distances away from zero the effects start to get bizarre and fall apart. The parenting method is sound but for example I am shooting through the glass representation of a vintage lens, six geometries of glass, and when I set the camera to something like y=-200 meters, the image completely falls apart.

Exact steps:

Model 6 pieces of glass (or plain refractive shader) for a lensing system at (0,0,0), and parent an orthographic camera to it.
Doesnt seem to matter, have seen same effects between perspective and my current method (orthographic the size of a 36mm piece of film with DOF turn on to smallest setting).
After parenting correct distances for a sharp image, move camera back something large from zero (like 200 meters for one of my products).

I can send an file if someone would like to take a look at what I am talking about. This happens with both the CPU and the GPU rendering for Cycles. I have a weird hunch a distance value during rendering is limited to a certain bit depth for the floating point value (either with rays or the refractive index material) and cannot compute the fine spacing between the glass elements correctly before bit rounding happens.

I understand this may be pushing the limits of Blender Cycles render but I am hoping to find a fix for the customers who have purchased other lensing systems from me as this limits their performance. This is affecting my anamorphic lens sets as well when at great distances. I have

OpticalIssue.blend

Thank you,

Jack

**System Information** Operating system: PC Graphics card: RTX3080Ti **Blender Version** Broken: 3.31, others Worked: none Hi, I am recreating the geometry of optical camera lenses to shoot through for photographic effect. I have noticed that the results look great with my current method when the camera is at location (0,0,0) however when moving to greater distances away from zero the effects start to get bizarre and fall apart. The parenting method is sound but for example I am shooting through the glass representation of a vintage lens, six geometries of glass, and when I set the camera to something like y=-200 meters, the image completely falls apart. Exact steps: 1. Model 6 pieces of glass (or plain refractive shader) for a lensing system at (0,0,0), and parent an orthographic camera to it. 2. Doesnt seem to matter, have seen same effects between perspective and my current method (orthographic the size of a 36mm piece of film with DOF turn on to smallest setting). 3. After parenting correct distances for a sharp image, move camera back something large from zero (like 200 meters for one of my products). I can send an file if someone would like to take a look at what I am talking about. This happens with both the CPU and the GPU rendering for Cycles. I have a weird hunch a distance value during rendering is limited to a certain bit depth for the floating point value (either with rays or the refractive index material) and cannot compute the fine spacing between the glass elements correctly before bit rounding happens. I understand this may be pushing the limits of Blender Cycles render but I am hoping to find a fix for the customers who have purchased other lensing systems from me as this limits their performance. This is affecting my anamorphic lens sets as well when at great distances. I have ![DistanceZero.png](https://archive.blender.org/developer/F14044126/DistanceZero.png) ![DistanceFarFromZero.png](https://archive.blender.org/developer/F14044128/DistanceFarFromZero.png) [OpticalIssue.blend](https://archive.blender.org/developer/F14044134/OpticalIssue.blend) Thank you, Jack

Jack Toth commented

2022-12-08 03:32:20 +01:00

Added subscriber: @Jack-Toth

Richard Antalik commented

2022-12-08 11:57:00 +01:00

Added subscriber: @iss

Richard Antalik commented

2022-12-08 11:57:00 +01:00

Changed status from 'Needs Triage' to: 'Needs User Info'

Richard Antalik commented

2022-12-08 11:57:00 +01:00

I can reproduce this behavior, seems that weird effects starts to happen at position greater than 32767m from 0, so this sounds like integer overflow.

In step 3 you say, to set position at 200m, but in provided file the distance is actually 300000m That is about orbit of the moon, so I would say, that is pushing limits a bit :) Is this correct or a mistake?

I can reproduce this behavior, seems that weird effects starts to happen at position greater than 32767m from 0, so this sounds like integer overflow. In step 3 you say, to set position at 200m, but in provided file the distance is actually 300000m That is about orbit of the moon, so I would say, that is pushing limits a bit :) Is this correct or a mistake?

Jack Toth commented

2022-12-08 14:54:51 +01:00

Thank you for taking a look. I had originally uploaded a file that showed it at 200m but changed to the single suzzanne model for simplicity where the effect happened at 300000m. I can prove this happens at even 6m with one of my lens stacks (where the effect is magnified because of the 6 lens elements) but I did not want to upload the actual product to a public forum, I can send privately if needed. My point was that this happens globally, being that the tearing is small but still effectful to a project like mine at even small distances from zero. The 300,000m just showcased the effect well to a normal viewing size. Is there a way to increase the resolution of position or is this constrained to 64-bit and can’t be fixed?

Therefore does mesh tearing, even at macro levels, always happen?

Thank you for taking a look. I had originally uploaded a file that showed it at 200m but changed to the single suzzanne model for simplicity where the effect happened at 300000m. I can prove this happens at even 6m with one of my lens stacks (where the effect is magnified because of the 6 lens elements) but I did not want to upload the actual product to a public forum, I can send privately if needed. My point was that this happens globally, being that the tearing is small but still effectful to a project like mine at even small distances from zero. The 300,000m just showcased the effect well to a normal viewing size. Is there a way to increase the resolution of position or is this constrained to 64-bit and can’t be fixed? Therefore does mesh tearing, even at macro levels, always happen?

Thomas Dinges commented

2023-01-20 16:18:33 +01:00

Changed status from 'Needs User Info' to: 'Needs Triage'

Brecht Van Lommel commented

2023-01-20 18:33:14 +01:00

Added subscriber: @brecht

Brecht Van Lommel commented

2023-01-20 18:33:14 +01:00

Changed status from 'Needs Triage' to: 'Archived'

Brecht Van Lommel closed this issue

2023-01-20 18:33:14 +01:00

Brecht Van Lommel commented

2023-01-20 18:33:14 +01:00

While ideally we could support such cases without precision issues, this is considered a known limitation.

The values here are below what we can reasonably do with single precision floating point values. A better solution would be some kind of camera shader or baked lens mapping to be used by Cycles for sampling rather than modelling it with geometry, but adding features like that is outside the scope of the bug tracker.

While ideally we could support such cases without precision issues, this is considered a known limitation. The values here are below what we can reasonably do with single precision floating point values. A better solution would be some kind of camera shader or baked lens mapping to be used by Cycles for sampling rather than modelling it with geometry, but adding features like that is outside the scope of the bug tracker.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Refractive Errrors when Camera is far from (0,0,0) #103005