EEVEE-Next: Add mesh volume bounds estimation #113731

Clément Foucault · 2023-10-14T23:24:22+02:00

Clément Foucault commented

2023-10-14 23:24:22 +02:00

This adds correct object bounds estimation.

This works by creating an occupancy texture where one
bit represents one froxel. A geometry pre-pass fill this
occupancy texture and doesn't do any shading. Each bit
set to 0 will not be considered occupied by the object
volume and will discard the material compute shader for
this froxel.

There is 2 method of computing the occupancy map:

Atomic XOR: For each fragment we compute the amount of
froxels center in-front of it. We then convert that
into occupancy bitmask that we apply to the occupancy
texture using imageAtomicXor. This is straight forward
and works well for any manifold geometry.
Hit List: For each fragment we write the fragment depth
in a list (contained in one array texture). This list
is then processed by a fullscreen pass (see
eevee_occupancy_convert_frag.glsl) that sorts and
converts all the hits to the occupancy bits. This
emulate Cycles behavior by considering only back-face
hits as exit events and front-face hits as entry events.
The result stores it to the occupancy texture using
bit-wise OR operation to compose it with other non-hit
list objects. This also decouple the hit-list evaluation
complexity from the material evaluation shader.

Limitations

Fast

Non-manifolds geometry objects are rendered incorrectly.
Non-manifolds geometry objects will affect other objects
in front of them.

Accurate

Limited to 16 hits per layer for now.
Non-manifolds geometry objects will affect other objects
in front of them.

Issues to fix

Crossing the surface with the camera near plane
make the mesh non-manifold and exhibit the
same issue. Fixed by simply reversing the XOR input.
This way missing front geometry fill the space between
the camera and the first surface depth.
Surface crossing camera far plane exhibit issues
(Caused by above fix). Fix is to use infinite projection
matrix to never clip.
Non-Manifold object affect other objects. Clip
occupancy bits to the bbox. This is a bit involved and
might disable some optimization. Consider a limitation
for now.
Object intersections are effectively removed
instead of overlapping. Use layered approach.
Volumes objects do not use this prepass making
them not working at all. Use selection surface mesh.
Volume object have multiple boxes in their coarse
geometry. This causes overlapping issues. Might be fixed
by drawing with front-face culling, limiting occupancy
to the cubes volumes and using atomicOr instead of
atomicXor. This would work because cubes are not
arbitrary geometry. We might want to do this for
point clouds too. NOTE: This was delayed and for now,
volumes still use their bounding boxes.
Make sure hit lists are allocated only when needed.

Follow ups

Object raster does not follow the froxels jitter
in XY dimensions (it does in Z). Needs to be fixed as
part of changing the sampling scheme.
Make it possible to have more than 16 hits.

Examples

Fast

Complex Mesh	Manifold Geometry	Overlapping volumes

	Note: The suzanne being non-manifold by default, this one was edited to make it watertight.	Note: Here suzanne is non-manifold. Note correct intersection.

Accurate

Cycles	EEVEE

This adds correct object bounds estimation. This works by creating an occupancy texture where one bit represents one froxel. A geometry pre-pass fill this occupancy texture and doesn't do any shading. Each bit set to 0 will not be considered occupied by the object volume and will discard the material compute shader for this froxel. There is 2 method of computing the occupancy map: - Atomic XOR: For each fragment we compute the amount of froxels **center** in-front of it. We then convert that into occupancy bitmask that we apply to the occupancy texture using `imageAtomicXor`. This is straight forward and works well for any manifold geometry. - Hit List: For each fragment we write the fragment depth in a list (contained in one array texture). This list is then processed by a fullscreen pass (see `eevee_occupancy_convert_frag.glsl`) that sorts and converts all the hits to the occupancy bits. This emulate Cycles behavior by considering only back-face hits as exit events and front-face hits as entry events. The result stores it to the occupancy texture using bit-wise `OR` operation to compose it with other non-hit list objects. This also decouple the hit-list evaluation complexity from the material evaluation shader. ## Limitations ### Fast - Non-manifolds geometry objects are rendered incorrectly. - Non-manifolds geometry objects will affect other objects in front of them. ### Accurate - Limited to 16 hits per layer for now. - Non-manifolds geometry objects will affect other objects in front of them. ## Issues to fix - [x] ~~Crossing the surface with the camera near plane make the mesh non-manifold and exhibit the same issue.~~ Fixed by simply reversing the XOR input. This way missing front geometry fill the space between the camera and the first surface depth. - [x] ~~Surface crossing camera far plane exhibit issues (Caused by above fix).~~ Fix is to use infinite projection matrix to never clip. - [x] ~~Non-Manifold object affect other objects. Clip occupancy bits to the bbox.~~ This is a bit involved and might disable some optimization. Consider a limitation for now. - [x] ~~Object intersections are effectively removed instead of overlapping.~~ Use layered approach. - [x] ~~Volumes objects do not use this prepass making them not working at all.~~ Use selection surface mesh. - [x] ~~Volume object have multiple boxes in their coarse geometry. This causes overlapping issues.~~ Might be fixed by drawing with front-face culling, limiting occupancy to the cubes volumes and using `atomicOr` instead of `atomicXor`. This would work because cubes are not arbitrary geometry. We might want to do this for point clouds too. NOTE: This was delayed and for now, volumes still use their bounding boxes. - [x] Make sure hit lists are allocated only when needed. ### Follow ups - Object raster does not follow the froxels jitter in XY dimensions (it does in Z). Needs to be fixed as part of changing the sampling scheme. - Make it possible to have more than 16 hits. ## Examples ### Fast | Complex Mesh | Manifold Geometry | Overlapping volumes | | -------- | -------- | -------- | | ![Capture d’écran du 2023-10-14 23-12-04](/attachments/b4f3011b-511e-4b25-894a-24373234aed3) | ![Capture d’écran du 2023-10-14 22-56-08](/attachments/0174b930-3ff2-4c09-98a0-59e7dd09f010) | ![image](/attachments/b22efd34-1964-4406-9dbf-d34f0c3bdf02) | | | Note: The suzanne being non-manifold by default, this one was edited to make it watertight. | Note: Here suzanne is non-manifold. Note correct intersection. | ### Accurate | Cycles | EEVEE | | -------- | -------- | |![Capture d’écran du 2023-10-17 21-30-37](/attachments/6b86af6f-f72c-4537-9535-44281c6da228)|![Capture d’écran du 2023-10-17 21-30-43](/attachments/2f8256fc-a7fc-415e-968d-9ef76f0c32b5)|

Capture d’écran du 2023-10-17 21-30-37.png

314 KiB

image.png

209 KiB

Capture d’écran du 2023-10-14 23-12-04.png

392 KiB

Capture d’écran du 2023-10-14 22-56-08.png

285 KiB

Capture d’écran du 2023-10-17 21-30-43.png

258 KiB

❤️ 3

Clément Foucault added this to the 4.1 milestone 2023-10-14 23:24:22 +02:00

Clément Foucault added the

 @ -295,13 +313,18 @@ Material &MaterialModule::material_sync(Object *ob,
         mat.planar_probe_shading = material_pass_get(
             ob, blender_mat, MAT_PIPE_DEFERRED, geometry_type, MAT_PROBE_PLANAR);
       }
     }

 @ -45,1 +47,4 @@
   MAT_GEOM_GPENCIL,
   MAT_GEOM_VOLUME,
   /* These maps to special vertex shader. */

 @ -768,1 +831,3 @@
   inst_.sampling.bind_resources(volume_ps_);
   enabled_ = false;
   for (auto &layer : layers_) {
     (*layer).sync();

 @ -769,0 +841,4 @@
   for (auto &layer : layers_) {
     /* TODO(fclem): We might want to skip empty layers as the clear overhead is be significant. */
     /* TODO(fclem): Move this clear inside the render pass. */
     occupancy_tx.clear(uint4(0u));

 @ -542,33 +644,29 @@ class PipelineModule {
       case MAT_PIPE_DEFERRED_PREPASS:
         return deferred.prepass_add(blender_mat, gpumat, false);
       case MAT_PIPE_FORWARD_PREPASS:
         if (GPU_material_flag_get(gpumat, GPU_MATFLAG_TRANSPARENT)) {

 @ -278,0 +280,4 @@
   GridAABB(int3 min_, int3 max_) : min(min_), max(max_){};
   /** Returns the intersection between this AABB and the \a other AABB. */
   GridAABB intersect(const GridAABB &other) const

 @ -441,0 +444,4 @@
                                            MAT_GEOM_VOLUME_WORLD,
                                            MAT_GEOM_VOLUME_OBJECT,
                                            MAT_GEOM_VOLUME,
                                            MAT_GEOM_VOLUME);

 @ -154,3 +156,3 @@
     GPUMaterial *gpu_material = material_array.gpu_materials[i];
     if (material.volume.gpumat && i == 0) {
     geometry_call(material.volume_occupancy.sub_pass, geom, res_handle);

 @ -315,0 +333,4 @@
   /* Use bounding box tag empty spaces. */
   GPUBatch *geom = DRW_cache_cube_get();
   geometry_call(material.volume_occupancy.sub_pass, geom, res_handle);

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

EEVEE-Next: Add mesh volume bounds estimation #113731

Limitations

Fast

Accurate

Issues to fix

Follow ups

Examples

Fast

Accurate