Skip to content

Troubleshooting

  1. I get a “CUDA out of memory” exception when rendering, what shall I do?

    Check if the models and scenarios you try to render can be rendered on your computer and do not exceed your GPU’s RAM limit. To test this, open the file in blender and try to render it from the GUI. Blender should indicate how much memory it requires for rendering. Compare this value with your GPU RAM.

  2. I get a “CUDA out of memory” exception, but I have a GPU with gazillion GB of RAM!

    There appears to be an issue for certain scenarios and models that have many light sources and different procedural textures. In this case, it happens every once in a while, that the BVH that is generated by blender does not fit into the GPU. It is currently unclear if this is a blender problem, a driver issue, or something unrelated.

    If you experience this issue, try to re-run rendering (both within blender, or the scripts), as this issue appears to happen only sporadically. If rendering starts from the script, it will finish - rendering and moving data to the GPU is performed only in the beginning.

    If re-running does not solve this issue, try to reduce the amount of light sources and models in your scenario.

  3. Why does forward simulating frames require lots of time?

    The reason for this is that, to drop objects physically correctly, we use blenders physics engine (in turn, this is bullet physics), and set the shape transform in blender to “Mesh”. The time the physics simulation takes depends crucially on the model you use, an in particular the number of vertices your model has. The rule of thumb is: fewer vertices are faster to simulate forward.

    Conclusively, either wait until it’s done, or use a low-poly model of your target object.

  4. I set up my custom sceneario and its python backend but when I try to run abrgen I get a “Runtime” error telling that ABR does not know my scenario!

    Most likely you forgot to let ABR automatically register your scenario by using abr_scenes.register(name=, type=). Please refer to our tutorial

    If the above is not the case, in your configuration .cfg file make sure that dataset.scene_type is set to be exactly equal to your selected _scene_name (see our tutorial).