workaround to deal with vk::Device::allocateDescriptorSetsUnique: ErrorOutOfPoolMemory

[keiohta]

Hi author, I have sometimes encountered an error

vk::Device::allocateDescriptorSetsUnique: ErrorOutOfPoolMemory

when taking a picture using a camera. In particular, this happens when I call camera.take_picture() multiple times (like 10K-100K times) to create a dataset. More concretely,

1. Create an engine
2. For around 1000 URDFs
   1. Create a scene and load an object via URDF
   2. For around 1000 times
      1. Randomize the initial pose of the object
      2. Take some pictures from different angles and save it
   3. Finish this loop without removing the object and the scene

Watching used memories for this script using htop and nvidia-smi, there are enough RAMs for both even when this error terminates the process. In general, do you have any idea when this error likely happens? Or do you know some way to release pool memory for Vulkan Renderer explicitly?

Keeping creating new scenes (and articulations under the scene) can be a problem, but it might not be related to this problem.

I understand this error is not directly related to SAPIEN, but it would be great if you could share your knowledge on this. Thanks in advance for your time.

[fbxiang]

You are probably hitting the maximum number of materials and textures. You can increase them by passing larger numbers to the VulkanRenderer constructor.

SAPIEN/python/pysapien_content.hpp

Lines 1736 to 1737 in 6bc3f4e

	py::arg("offscreen_only") = false, py::arg("max_num_materials") = 5000,
	py::arg("max_num_textures") = 5000, py::arg("default_mipmap_levels") = 1,

[keiohta]

This perfectly solved the problem I had been struggling with for six months. Thanks a lot!!!