[Bug] Benchmarking tool running immediately and not producing any output

[davidlauerman]

I am having issues running the benchmarking tool, I am not sure if this is something that has been encountered. In the terminal, I am inputting :
python tools/benchmarking/benchmark.py --config tools/benchmarking/benchmark_params.yaml
and receiving as an output almost immediately (about 2 seconds in total):
Benchmarking started 🏃‍♂️. This will take a while ⏲ depending on your configuration. 2023-06-08 15:31:28,554 - /app/myanomalib/anomalib/tools/benchmarking/benchmark.py - WARNING - Config requested GPU benchmarking but torch could not detect any cuda enabled devices Finished gathering results ⚡
This is my benchmark_params.yaml file content:

seed: 42
compute_openvino: false
hardware:
  - gpu
writer:
  - wandb
grid_search:
  dataset:
    name: hazelnut
    format: folder
    root: {str(root_directory / "datasets" / "hazelnut_toy")}
    normal_dir: good # name of the folder containing normal images.
    abnormal_dir: colour # name of the folder containing abnormal images.
    normal_test_dir: null
    task: segmentation # classification or segmentation
    mask: /anomalib/datasets/hazelnut_toy/mask/colour
    extensions: .jpg
    split_ratio: 0.2
    category:
      - colour
      - crack
    image_size: [128, 256]
  model_name:
    - padim
    - stfpm

I am just hoping to get some sort of output, but as of right now there is nothing and so I am not sure where to start. I have not changed anything else outside of this file, I have been just working my way through the 'Training with Custom Data' tutorial. I am running this all in a docker container that is almost identical to the developer container in the git repository. Let me know if anything seems familiar here or anything jumps out. Thanks!

[blaz-r]

Hello @davidlauerman.
The error says you don't have cuda enabled gpus on your system. Can you verify that you have proper version of torch, that also includes cuda support?

[blaz-r]

Even after this. Your config doesn't look quite right. Currently there is a discrepancy between docs, code and notebooks, so this should be fixed. The same issue also occurred in #1109. I believe @ashwinvaidya17, as the author of benchmark, knows the most about this.

The default included benchmark_params.yaml does work, if you want to use mvtec examples.

Here is also config you have for hazelnut_toy that I fixed and verified that is working:

seed: 42
compute_openvino: false
hardware:
  - gpu
writer:
  - wandb
grid_search:
  dataset:
    name: hazelnut
    format: folder
    root: datasets/hazelnut_toy
    normal_dir: good # name of the folder containing normal images.
    abnormal_dir: colour # name of the folder containing abnormal images.
    normal_test_dir: null
    task: segmentation # classification or segmentation
    mask_dir: mask/colour
    extensions: .jpg
    split_ratio: 0.2
    category:
      - colour
      - crack
    image_size: [128, 256]
  model_name:
    - padim
    - stfpm