Error: Not freed memory blocks...

[freli19]

Describe the issue

Hello,
I am trying to run a simple script that is like a hybrid of examples/basic/physics_positioning and examples/advanced/point_clouds. After running it a number of times, bproc.object.simulate_physics_and_fix_final_poses() takes longer and longer to run, and when finished I get the print out:
Blender quit
Error: Not freed memory blocks: 9, total unfreed memory 49.438984 MB
Cleaning temporary directory
What can be done to delete or free these memory blocks? I have tried to find the solution but I am yet to succeed. And also, is there a common cause for the accumulation of these?
Thank you!

Minimal code example

bproc.init()
# load blender materials from cubes
blendermetallic = bproc.loader.load_blend("U:/Data\Objekt\Blender/metallic_cube.blend", data_blocks='materials')
# Load the chosen object into the scene
active = []
for i in range(args.num_objects):
    active.append(bproc.loader.load_obj("u:\Data\object.obj")[0])

# load a surface
scene = [bproc.loader.load_obj("U:\Data\plane.obj")[0]]
scene_pos = np.array([0, 0, -1500]) # pose in mm
scene[0].set_location(scene_pos)

# load a bin for objects to fall in to
scene.append(bproc.loader.load_obj("U:\Data/bin.obj")[0])
scene[1].set_location(scene_pos + [0, 0, 110]) # pose in mm

# Make all objects actively participate in the simulation
for obj in active:
    obj.enable_rigidbody(active=True)
    obj.add_material(metallic[1])
    obj.set_cp("category_id", 1)

# The scene should only act as an obstacle and is therefore marked passive.
# To let the objects fall into the valleys of the ground, make the collision shape MESH instead of CONVEX_HULL.
for obj in scene:
    obj.enable_rigidbody(active=False, collision_shape="MESH")

# Define a function that samples the pose of a given sphere
def sample_pose(obj: bproc.types.MeshObject):
    obj.set_location(np.random.uniform(scene_pos + [-150, -150, 300], scene_pos + [150, 150, 500]))
    print(obj.get_location())
    obj.set_rotation_euler(bproc.sampler.uniformSO3())
    
# Sample the poses of all objects above the ground without any collisions in-between
bproc.object.sample_poses(
    active,
    sample_pose_func=sample_pose
)

# Run the simulation and fix the poses of the objects at the end
bproc.object.simulate_physics_and_fix_final_poses(min_simulation_time=4, max_simulation_time=60, check_object_interval=1)

bvh_tree = bproc.object.create_bvh_tree_multi_objects(active + scene)

# define the camera resolution for the pc
bproc.camera.set_resolution(1024, 1024)

bproc.camera.set_intrinsics_from_blender_params(clip_start=1200.0, clip_end=1600.0)
position, euler_rotation = [0, 0, 0], [0, 0, 0]
matrix_world = bproc.math.build_transformation_mat(position, euler_rotation)
bproc.camera.add_camera_pose(matrix_world)

# Compute a depth image from the view of the second camera pose
depth = bproc.camera.depth_via_raytracing(bvh_tree, 1)

# Project the depth again to get a point cloud
points = bproc.camera.pointcloud_from_depth(depth, 1)
points = points.reshape(-1, 3)

bproc.renderer.enable_depth_output(activate_antialiasing=False)

# set camera res for rendering
bproc.camera.set_resolution(512, 512)

# render the whole pipeline
data = bproc.renderer.render()
# Write BOP-data
bproc.writer.write_bop(output_dir = "./output/",
                       target_objects = active,
                       depths = data["depth"],
                       colors = data["colors"],
                       color_file_format = "PNG",
                       append_to_existing_output=True,
                       save_world2cam=True,
                       ignore_dist_thres=3000.0,
                       annotation_unit='m'
                       )

Files required to run the code

No response

Expected behavior

I expect the simulation to finish within 20 seconds as it did before. Now i get the following prints:

BlenderProc version

2.8.0

[cornerfarmer]

Hi @freli19,

sorry for the late response. Were you able to solve your issue in the meantime?
Are you running your code in a loop or do you do separate script runs?

[freli19]

Hello,

No worries! I was not able to solve this problem yet, but it has not stopped me from using BlenderProc the last month. Since posting, the total unfreed memory did mysteriously decrease (and stayed at around 9 MB for many runs) only to increase again yesterday and today... I run separate script runs, it is not run in a loop.

One theory I had for the decrease in unfreed memory was that I changed
obj.add_material(metallic[1])
to
obj.set_material(0, metallic[1]),
but since it now increased again, I am no longer sure what the problem is. The simulation is already slow because of the scale, but I believe the scale of the objects should not explain the increase in unfreed memory and simulation time from run to run. Also, I still do not know how to free this unfreed memory.

[cornerfarmer]

If you are using separate script runs, then there should be no memory sharing between them and each script run should behave the same. Are sure that the physics simulation is getting slower and slower? Is this maybe caused by using different objects each time?

[freli19]

Hm, I was pretty sure that the physics simulation was taking increasingly more time for each run back in early December. But now I am no longer able to reproduce these results.
Now I simply get a very linear increase of unfreed memory when I increase the number of objects included in the simulation but when I decrease the number of objects, the unfreed memory is also decreased. I get the occasional really slow simulation, but now it seems this is due to the properties of the scene and the randomness of the pose sampler.
So if the print Error: Not freed memory blocks: 9, total unfreed memory 39.556110 MB is nothing to worry about, I guess this issue can be closed. Thank you for answers and for an amazing pipeline!