How to use custom transforms for training and evaluation

[letmejoin]

What is the motivation for this task?

1. new version anomalib reduce transforms from dataset Calss, but I don't konw how to add it with config.yaml
2. The error message do not show enough information for solving
3. 

Describe the solution you'd like

Pleace give usage for customed data preprocessing, such as CenterCrop, Normalize

Additional context

No response

[samet-akcay]

Can you share your config file here?

[letmejoin]

That's my config,thanks!
seed_everything: true
trainer:
accelerator: auto
strategy: auto
devices: 1
num_nodes: 1
precision: 32
logger: null
callbacks: null
fast_dev_run: false
max_epochs: null
min_epochs: null
max_steps: -1
min_steps: null
max_time: null
limit_train_batches: null
limit_val_batches: null
limit_test_batches: null
limit_predict_batches: null
overfit_batches: 0.0
val_check_interval: null
check_val_every_n_epoch: 1
num_sanity_val_steps: null
log_every_n_steps: null
enable_checkpointing: null
enable_progress_bar: null
enable_model_summary: null
accumulate_grad_batches: 1
gradient_clip_val: null
gradient_clip_algorithm: null
deterministic: null
benchmark: null
inference_mode: true
use_distributed_sampler: true
profiler: null
detect_anomaly: false
barebones: false
plugins: null
sync_batchnorm: true
reload_dataloaders_every_n_epochs: 0
default_root_dir: patchcore_train
normalization:
normalization_method: MIN_MAX

task: CLASSIFICATION
metrics:
image:

• AUROC
• F1Score
  pixel: null
  threshold:
  class_path: anomalib.metrics.F1AdaptiveThreshold
  init_args:
  default_value: 0.5
  thresholds: null
  validate_args: true
  compute_on_cpu: false
  dist_sync_on_step: false
  sync_on_compute: true
  logging:
  log_graph: false
  default_root_dir: 'resutls/diamond_andriod'
  ckpt_path: null
  data:
  class_path: anomalib.data.Folder
  init_args:
  name: iphone
  normal_dir: train/good
  root: datasets/Diamond_andriod
  abnormal_dir: test/bad
  normal_test_dir: test/good
  mask_dir: null
  normal_split_ratio: 0.0
  extensions: [".jpeg", ".jpg"]
  image_size:
  • 1280 # dimensions to which images are resized (mandatory)
  • 1280
    train_batch_size: 4
    eval_batch_size: 1
    num_workers: 10
    task: CLASSIFICATION
    transform:
    class_path: torchvision.transforms.v2.Normalize
    init_args:
    mean: [0.485, 0.456, 0.406]
    std: [0.229, 0.224, 0.225]
    class_path: torchvision.transforms.v2.CenterCrop
    init_args:
    size:
    • 448
    • 448
      test_split_mode: from_dir
      test_split_ratio: 0.2
      val_split_mode: same_as_test
      val_split_ratio: 0.5
      seed: null

model:
class_path: anomalib.models.Patchcore
init_args:
backbone: tf_efficientnet_b6_ns
layers:
- blocks.2.5
- blocks.4.7
pre_trained: true
coreset_sampling_ratio: 0.1
num_neighbors: 9

the rest of my config which commented:
visualization:
class_path: BaseVisualizer
visualizers:
class_path: ImageVisualizer
init_args:
mode: FULL
task: CLASSIFICATION
save: True
log: True
show: False

[samet-akcay]

This documentation shows how to enable custom transforms via both API and CLI
https://anomalib.readthedocs.io/en/latest/markdown/guides/how_to/data/transforms.html

[letmejoin]

Thanks for your replay！I get it!
But for other usage, I still don't know how to setup.
For example, how to modify the saving path of inference result images? The current image saving path is nested very deeply; the configuration parameters of visualization and visualizers seem to have been removed from the previous version; the specific code of the export model is also missing. How to configure it? The parameters of the exported model are not introduced in detail.
I feel that a detailed documentation is needed for each configuration parameter and configuration example of the configuration parameter, just like the configuration example of transform you gave me, but I didn't know where to find this introduction before.
I hope you can understand my explanation and be more friendly to developers who are new to this platform. Thank you very much!