Can this run on Ubuntu?

[astewzxc]

Hi, thanks to this great work.
I've run Anomalib on Windows 10 PC, and got good outcome.
However, when I try to run it on Ubuntu, I found the number of model parameters are 24.9M(68.9M on Windows),and got significant lower performance compare to that on Windows.

Ubuntu 20.04
NVIDIA CUDA 11.6.0
PyTorch 1.12.0
TensorRT 8.2.3

[samet-akcay]

Hi @astewzxc, can you provide more information to be able to reproduce the issue? For example, your config file, the CLI command you ran etc?

[astewzxc]

I use a cloud computing service from here
The CLI command I ran are:
pip install anomalib
python tools/train.py --model patchcore

but got an error:
ImportError: libGL.so.1: cannot open shared object file: No such file or directory

so I ran:
pip uninstall opencv-python
pip install opencv-python-headless

then the error resolved, so I ran
python tools/train.py --model patchcore

again. It worked, but the model parameters are much less than it ran on Windows.
About the config file, I didn't change anything yet.

[astewzxc]

The other thing I forgot to mentioned is that the version of Amomalib I run on Ubuntu is newer than on Windows.
If I ran the older version(version ran on Windows), an error occured:
omegaconf.errors.ConfigAttributeError: Missing key pre_trained

[astewzxc]

The config file of the older version is below:

dataset:
name: mvtec #options: [mvtec, btech, folder]
format: mvtec
path: ./datasets/MVTec
task: segmentation
category: grid
image_size: 224
train_batch_size: 32
test_batch_size: 1
num_workers: 8
transform_config:
train: null
val: null
create_validation_set: false
tiling:
apply: false
tile_size: null
stride: null
remove_border_count: 0
use_random_tiling: False
random_tile_count: 16

model:
name: patchcore
backbone: wide_resnet50_2
layers:
- layer2
- layer3
coreset_sampling_ratio: 0.1
num_neighbors: 9
weight_file: weights/model.ckpt
normalization_method: min_max # options: [null, min_max, cdf]

metrics:
image:
- F1Score
- AUROC
pixel:
- F1Score
- AUROC
threshold:
image_default: 0
pixel_default: 0
adaptive: true

project:
seed: 0
path: ./results

logging:
log_images_to: ["local"] # options: [wandb, tensorboard, local]. Make sure you also set logger with using wandb or tensorboard.
logger: [] # options: [tensorboard, wandb, csv] or combinations.
log_graph: false # Logs the model graph to respective logger.

PL Trainer Args. Don't add extra parameter here.

trainer:
accelerator: auto # <"cpu", "gpu", "tpu", "ipu", "hpu", "auto">
accumulate_grad_batches: 1
amp_backend: native
auto_lr_find: false
auto_scale_batch_size: false
auto_select_gpus: false
benchmark: false
check_val_every_n_epoch: 1 # Don't validate before extracting features.
default_root_dir: null
detect_anomaly: false
deterministic: false
devices: 1
enable_checkpointing: true
enable_model_summary: true
enable_progress_bar: true
fast_dev_run: false
gpus: null # Set automatically
gradient_clip_val: 0
ipus: null
limit_predict_batches: 1.0
limit_test_batches: 1.0
limit_train_batches: 1.0
limit_val_batches: 1.0
log_every_n_steps: 50
log_gpu_memory: null
max_epochs: 1
max_steps: -1
max_time: null
min_epochs: null
min_steps: null
move_metrics_to_cpu: false
multiple_trainloader_mode: max_size_cycle
num_nodes: 1
num_processes: null
num_sanity_val_steps: 0
overfit_batches: 0.0
plugins: null
precision: 32
profiler: null
reload_dataloaders_every_n_epochs: 0
replace_sampler_ddp: true
strategy: null
sync_batchnorm: false
tpu_cores: null
track_grad_norm: -1
val_check_interval: 1.0 # Don't validate before extracting features.

[astewzxc]

I found out it might be the version problem. I ran v0.3.2 on Ubuntu and got good results.

[samet-akcay]

Thanks for the update @astewzxc. I'm moving this to discussions now.