Compositor: Viewport issues when zoomed in #111344

New Issue

Open

opened 2023-08-21 14:55:25 +02:00 by Dalai Felinto · 4 comments

Dalai Felinto commented

2023-08-21 14:55:25 +02:00

Owner

Blender Version
Broken: version: 4.0.0 Alpha, branch: main, commit date: 2023-08-21 10:12, hash: f0a8c1946162
Worked: never

Short description of error
Composition Mask is "rendered" wrongly in the viewport when the camera is not fully in view.

Exact steps for others to reproduce the error

Open this file: zoomed_in.blend
Notice that in the viewport the view is zoomed in inside the camera view.

Current buggy result:

The mask only matches when the entire camera is in view:

**Blender Version** Broken: version: 4.0.0 Alpha, branch: main, commit date: 2023-08-21 10:12, hash: `f0a8c1946162` Worked: never **Short description of error** Composition Mask is "rendered" wrongly in the viewport when the camera is not fully in view. **Exact steps for others to reproduce the error** * Open this file: [zoomed_in.blend](/attachments/6e786259-47d2-4d0b-b8b4-96dc8a51089f) * Notice that in the viewport the view is zoomed in inside the camera view. Current buggy result: ![image](/attachments/3d03ef99-cca0-4c22-b02f-e6d8219ba5c9) The mask only matches when the entire camera is in view: ![image](/attachments/afa94e35-ecc5-40af-a4df-c340c4ca3f20)

zoomed_in.blend

130 KiB

148 KiB

100 KiB

Dalai Felinto added the

labels 2023-08-21 14:55:47 +02:00

Dalai Felinto added this to the Compositing project 2023-08-21 14:55:53 +02:00

Dalai Felinto referenced this issue

2023-08-21 15:00:56 +02:00

Compositor: Viewport issues when Passepartout is off #111345

Germano Cavalcante added

and removed

labels 2023-08-21 19:07:59 +02:00

Germano Cavalcante commented

2023-08-21 19:09:44 +02:00

Member

I can confirm the problem.
I suppose it has something to do with the rcti get_compositing_region() function in compositor_engine.cc.

The code was introduced in cc78fd4e93 by @OmarEmaraDev

I can confirm the problem. I suppose it has something to do with the `rcti get_compositing_region()` function in `compositor_engine.cc`. The code was introduced in cc78fd4e9364338126e5145ff96ae6c544939042 by @OmarEmaraDev

Germano Cavalcante added the

EEVEE & Viewport

label 2023-08-21 19:30:16 +02:00

Omar Emara commented

2023-08-22 14:25:06 +02:00

Member

Consider the following example where we have a 1000x800 viewport and a 500x500
final render resolution, and consequently a camera of the same resolution.

We have two options for how the compositor should operate on that setup, each of
which is discussed in the following two sections.

Camera Compositing

The compositor can constrain itself to the camera region, and out of camera
regions will be left unaltered regardless of passepartout.

Advantages:

The aspect ratio of the compositing space will match that of the final render
compositing pipeline.
Input resources that assume the size of the compositing space like Masks will
have the same size in both compositions and will align with the camera in their
identity transform.
The result of compositing will be viewport size invariant and the user will be
able to align images easily.

Disadvantages:

Regions outside of the camera will remain unaltered, which can look weird if
the passepartout is not opaque.

Viewport Compositing

The compositor will operate on the entire viewport and will ignore the camera
view.

The aforementioned advantages and disadvantages are essentially reversed, but
can be partially alleviated using the alternatives described in each of the
following sections.

Transformed Input Resources

Input resources can assume the size of the camera region and transformed
relative to the space of viewport such that they align with the camera.

The disadvantages is that pre-transformed resources can be unintuitive to work
with, for instance, rotating the resource will also move it due to the centered
pivot of the compositor transforms. Moreover, when realized on the domain of the
viewport, the areas outside of the original mask will assume a zero values,
which might not be what the users wants. Finally, the different domains of the
inputs can be harder to reason about for users, as opposed to having all inputs
have the same domain.

Padded Input Resources

Input resources can be zero padded to the size of the viewport, such that all
inputs will have the same size.

This alleviates the different sizes disadvantages but shares the rest of the
disadvantages as the transformed input resources alternative.

Neither of the alternatives workarounds the aspect ratio change and the
difference in sizes. To workaround those, one would have to change all nodes to
operate in the scene camera space. Except this would break setups where one is
processing an image that is standalone or simply is not related to the scene.

Consider the following example where we have a 1000x800 viewport and a 500x500 final render resolution, and consequently a camera of the same resolution. ![Illustration](/attachments/bc12154b-96c3-478e-a4ce-c59ef083dcdb) We have two options for how the compositor should operate on that setup, each of which is discussed in the following two sections. ## Camera Compositing The compositor can constrain itself to the camera region, and out of camera regions will be left unaltered regardless of passepartout. #### Advantages: - The aspect ratio of the compositing space will match that of the final render compositing pipeline. - Input resources that assume the size of the compositing space like Masks will have the same size in both compositions and will align with the camera in their identity transform. - The result of compositing will be viewport size invariant and the user will be able to align images easily. #### Disadvantages: - Regions outside of the camera will remain unaltered, which can look weird if the passepartout is not opaque. ## Viewport Compositing The compositor will operate on the entire viewport and will ignore the camera view. The aforementioned advantages and disadvantages are essentially reversed, but can be partially alleviated using the alternatives described in each of the following sections. #### Transformed Input Resources Input resources can assume the size of the camera region and transformed relative to the space of viewport such that they align with the camera. The disadvantages is that pre-transformed resources can be unintuitive to work with, for instance, rotating the resource will also move it due to the centered pivot of the compositor transforms. Moreover, when realized on the domain of the viewport, the areas outside of the original mask will assume a zero values, which might not be what the users wants. Finally, the different domains of the inputs can be harder to reason about for users, as opposed to having all inputs have the same domain. #### Padded Input Resources Input resources can be zero padded to the size of the viewport, such that all inputs will have the same size. This alleviates the different sizes disadvantages but shares the rest of the disadvantages as the transformed input resources alternative. ---- Neither of the alternatives workarounds the aspect ratio change and the difference in sizes. To workaround those, one would have to change all nodes to operate in the scene camera space. Except this would break setups where one is processing an image that is standalone or simply is not related to the scene.

40 KiB

Germano Cavalcante commented

2023-08-22 15:39:14 +02:00

Member

Perhaps this deserves a topic on DevTalk.

For "Camera Compositing", in my opinion, even with the weirdness, I feel it would be more useful to constrain to the camera region.
(It would be interesting to see pictures showing the disadvantages.)

For "Viewport Compositing", if the camera aspect is kept for the composition region, I imagine the composition should also be cropped just like in camera view.
(Some kind of border should be drawn in this case.)

If compositing the entire viewport is really useful for some cases. Then it could have an option in the popover.
Like [X] Keep Rendering Aspect.

Perhaps this deserves a topic on [DevTalk](https://devtalk.blender.org/). For "Camera Compositing", in my opinion, even with the weirdness, I feel it would be more useful to constrain to the camera region. (It would be interesting to see pictures showing the disadvantages.) For "Viewport Compositing", if the camera aspect is kept for the composition region, I imagine the composition should also be cropped just like in camera view. (Some kind of border should be drawn in this case.) If compositing the entire viewport is really useful for some cases. Then it could have an option in the popover. Like `[X] Keep Rendering Aspect`.

Dalai Felinto commented

2023-08-24 10:50:18 +02:00

Author

Owner

@OmarEmaraDev It may help to clarify how people expect (or are expected) to use the Viewport Compositing (as oppose to the camera compositing).

The Camera Compositing doesn't need all nodes to be in Scene Camera Space. What it needs is for nodes to have their absolute pixels to be internally converted to the scene camera space relative to the camera 100% size.

And on top of that to have the aspect ratio respected. For me the simpler example is how border render should just work with the viewport compositor, and give you the same result as if you were to render it out with F12.

@OmarEmaraDev It may help to clarify how people expect (or are expected) to use the Viewport Compositing (as oppose to the camera compositing). --- The Camera Compositing doesn't need all nodes to be in Scene Camera Space. What it needs is for nodes to have their absolute pixels to be internally converted to the scene camera space relative to the camera 100% size. And on top of that to have the aspect ratio respected. For me the simpler example is how border render should just work with the viewport compositor, and give you the same result as if you were to render it out with F12.

Sergey Sharybin removed this from the Compositing project 2023-09-06 10:37:50 +02:00

Sign in to join this conversation.

No Label

Animation & Rigging

Asset Browser Project Overview

Automated Testing

Blender Asset Bundle

Dependency Graph

Development Management

EEVEE & Viewport

Images & Movies

Motion Tracking

Nodes & Physics

Pipeline, Assets & IO

Platforms, Builds & Tests

Render & Cycles

Render Pipeline

Sculpt, Paint & Texture

Video Sequencer

Virtual Reality

Blender 2.8 Project

Milestone 1: Basic, Local Asset Browser

Good First Issue

Animation & Rigging

Development Management

EEVEE & Viewport

Nodes & Physics

Pipeline, Assets & IO

Platforms, Builds & Tests

Render & Cycles

Sculpt, Paint & Texture

Needs Info from Developers

Needs Information from User

No Milestone

No project

No Assignees

3 Participants

Notifications

Due Date

The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#111344

No description provided.