Use Docker

Check NVIDIA Installation

nvidia-smi

You can see the driver version and cuda version, on my computer:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.57.02    Driver Version: 470.57.02    CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:01:00.0  On |                  N/A |
| 40%   39C    P5    21W / 160W |    833MiB /  5932MiB |      1%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

Check Docker

systemctl status docker.service
systemctl status docker.socket

Installation guide: https://docs.docker.com/engine/install/

Check NVIDIA Docker

sudo docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi

Installation guide: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker

Enable CUDA during Docker building

edit the /etc/docker/daemon.json with content:

{
    "runtimes": {
        "nvidia": {
            "path": "/usr/bin/nvidia-container-runtime",
            "runtimeArgs": []
        }
    },
    "default-runtime": "nvidia"
}

restart docker daemon:

sudo systemctl restart docker

Reference: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker

Build Docker

cd docker
./build

Run Docker

You can use docker like a CLI tool:

./run --help
./run --output_dir=$(pwd)/deep_tote

usage: blender [-h] [--output_dir OUTPUT_DIR] [--camera_type CAMERA_TYPE]
               [--camera_height CAMERA_HEIGHT] [--obstruction OBSTRUCTION]
               [--tote_length TOTE_LENGTH] [--tote_width TOTE_WIDTH]
               [--tote_height TOTE_HEIGHT] [--tote_thickness TOTE_THICKNESS]
               [--model_path MODEL_PATH]
               [--model_max_dimension MODEL_MAX_DIMENSION]
               [--max_faces MAX_FACES] [--num_regen NUM_REGEN]
               [--num_begin NUM_BEGIN] [--num_end NUM_END]
               [--num_pick NUM_PICK] [--max_bounces MAX_BOUNCES]
               [--samples SAMPLES] [--substeps_per_frame SUBSTEPS_PER_FRAME]
               [--enable_perfect_depth] [--enable_instance_segmap]
               [--enable_object_masks] [--enable_class_segmap]
               [--enable_mesh_info]

optional arguments:
  -h, --help            show this help message and exit
  --output_dir OUTPUT_DIR
                        output directory, the folder will be automatically
                        created if not exist
  --camera_type CAMERA_TYPE
                        different cameras have different fov and aspect ratio:
                        Photoneo-M | Photoneo-L | XYZ-SL(default)
  --camera_height CAMERA_HEIGHT
                        camera height in meter, default: 2
  --obstruction OBSTRUCTION
                        control the number of obstructed points, reasonable
                        range 0~0.4, default: 0.2
  --tote_length TOTE_LENGTH
                        tote x-axis dimension, default: 0.7
  --tote_width TOTE_WIDTH
                        tote y-axis dimension, default: 0.7
  --tote_height TOTE_HEIGHT
                        tote z-axis dimension, default: 0.5
  --tote_thickness TOTE_THICKNESS
                        tote thickness, default: 0.01
  --model_path MODEL_PATH
                        CAD model, supported format: ply, stl
  --model_max_dimension MODEL_MAX_DIMENSION
                        rescale the model to this value, if this value <= 0,
                        do nothing
  --max_faces MAX_FACES
                        decimate mesh if the number of faces of mesh is bigger
                        than this value, default: 10000
  --num_regen NUM_REGEN
                        run generation process for num_gen times, default: 2
  --num_begin NUM_BEGIN
                        number of objects at the beginning, default: 30
  --num_end NUM_END     number of objects in the end, default: 0
  --num_pick NUM_PICK   number of objects picked each time, default: 5
  --max_bounces MAX_BOUNCES
                        render option: max bounces of light, default: 3
  --samples SAMPLES     render option: samples for each pixel, default: 10
  --substeps_per_frame SUBSTEPS_PER_FRAME
                        physics option: higher value for higher simulation
                        stability, default: 10
  --enable_perfect_depth
                        flag: render depth without obstruction
  --enable_instance_segmap
                        flag: render instance segmentation map
  --enable_object_masks
                        flag: render object masks
  --enable_class_segmap
                        flag: render class segmentation map