quadra 0.0.1__py3-none-any.whl → 2.2.7__py3-none-any.whl

quadra/configs/experiment/generic/mvtec/anomaly/fastflow.yaml

@@ -0,0 +1,29 @@
+# @package _global_
+defaults:
+  - base/anomaly/fastflow
+  - override /datamodule: generic/mvtec/anomaly/base
+
+datamodule:
+  num_workers: 12
+  train_batch_size: 32
+  task: ${model.dataset.task}
+  category: bottle
+
+callbacks:
+  score_normalization:
+    threshold_type: "pixel"
+print_config: false
+
+core:
+  tag: "run"
+  test_after_training: true
+  upload_artifacts: true
+  name: fastflow_${datamodule.category}_${model.model.backbone}_${trainer.max_epochs}
+
+logger:
+  mlflow:
+    experiment_name: mvtec-anomaly
+    run_name: ${core.name}
+
+trainer:
+  devices: [2]

quadra/configs/experiment/generic/mvtec/anomaly/inference.yaml

@@ -0,0 +1,27 @@
+# @package _global_
+defaults:
+  - base/anomaly/inference
+  - override /datamodule: generic/mvtec/anomaly/base
+
+datamodule:
+  num_workers: 12
+  test_batch_size: 32
+  task: segmentation
+  category: bottle
+
+core:
+  tag: "run"
+  test_after_training: true
+  upload_artifacts: true
+  name: inference_${datamodule.category}
+
+logger:
+  mlflow:
+    experiment_name: mvtec-anomaly
+    run_name: ${core.name}
+
+trainer:
+  devices: [2]
+
+task:
+  model_path: ???

quadra/configs/experiment/generic/mvtec/anomaly/padim.yaml

@@ -0,0 +1,37 @@
+# @package _global_
+defaults:
+  - base/anomaly/padim
+  - override /datamodule: generic/mvtec/anomaly/base
+
+model:
+  model:
+    input_size: [224, 224]
+    backbone: resnet18.tv_in1k
+
+datamodule:
+  num_workers: 12
+  train_batch_size: 32
+  task: ${model.dataset.task}
+  category: bottle
+
+callbacks:
+  score_normalization:
+    threshold_type: "pixel"
+
+print_config: false
+
+core:
+  tag: "run"
+  test_after_training: true
+  upload_artifacts: true
+  name: padim_${datamodule.category}_${model.model.backbone}_${trainer.max_epochs}
+
+logger:
+  mlflow:
+    experiment_name: mvtec-anomaly
+    run_name: ${core.name}
+
+trainer:
+  devices: [2]
+  max_epochs: 1
+  check_val_every_n_epoch: ${trainer.max_epochs}

quadra/configs/experiment/generic/mvtec/anomaly/patchcore.yaml

@@ -0,0 +1,37 @@
+# @package _global_
+defaults:
+  - base/anomaly/patchcore
+  - override /datamodule: generic/mvtec/anomaly/base
+
+model:
+  model:
+    input_size: [224, 224]
+    backbone: resnet18.tv_in1k
+
+datamodule:
+  num_workers: 12
+  train_batch_size: 32
+  task: ${model.dataset.task}
+  category: bottle
+
+callbacks:
+  score_normalization:
+    threshold_type: "pixel"
+
+print_config: false
+
+core:
+  tag: "run"
+  test_after_training: true
+  upload_artifacts: true
+  name: patchcore_${datamodule.category}_${model.model.backbone}_${trainer.max_epochs}
+
+logger:
+  mlflow:
+    experiment_name: mvtec-anomaly
+    run_name: ${core.name}
+
+trainer:
+  devices: [2]
+  max_epochs: 1
+  check_val_every_n_epoch: ${trainer.max_epochs}

quadra/configs/experiment/generic/oxford_pet/segmentation/smp.yaml

@@ -0,0 +1,27 @@
+# @package _global_
+defaults:
+  - base/segmentation/smp # use smp file as default
+  - override /datamodule: generic/oxford_pet/segmentation/base # update datamodule
+  - override /loss: smp_dice_multiclass
+  - override /model: smp_multiclass
+  - _self_ # use this file as final config
+
+trainer:
+  devices: [0]
+  max_epochs: 10
+
+backbone:
+  model:
+    num_classes: 2 # The total number of classes (background + foreground)
+
+task:
+  report: true
+  evaluate:
+    analysis: true
+
+core:
+  save_for_runtime: False
+  name: "oxford_pet_segmentation" # change the name of experiment (also for Mlflow)
+
+datamodule:
+  batch_size: 64

quadra/configs/export/default.yaml

@@ -0,0 +1,13 @@
+types: [torchscript]
+input_shapes: # Redefine the input shape if not automatically inferred
+onnx:
+  # torch.onnx.export options
+  input_names: # If null automatically inferred
+  output_names: # If null automatically inferred
+  dynamic_axes: # If null automatically inferred
+  export_params: true
+  opset_version: 16
+  do_constant_folding: true
+  # Custom options
+  fixed_batch_size: # If not null export with fixed batch size (ignore dynamic axes)
+  simplify: true

quadra/configs/hydra/anomaly_custom.yaml

@@ -0,0 +1,15 @@
+# @package _global_
+
+hydra:
+  run:
+    dir: logs/runs/${core.name}/${now:%Y-%m-%d_%H-%M-%S}
+  sweep:
+    dir: logs/multiruns/${now:%Y-%m-%d_%H-%M-%S}/
+    subdir: ${multirun_subdir_beautify:${hydra.job.override_dirname}}
+  job:
+    config:
+      override_dirname:
+        exclude_keys:
+          - trainer.devices
+          - logger.mlflow.experiment_name
+    chdir: true

quadra/configs/hydra/default.yaml

@@ -0,0 +1,14 @@
+# @package _global_
+
+hydra:
+  run:
+    dir: logs/runs/${core.name}/${now:%Y-%m-%d_%H-%M-%S}
+  sweep:
+    dir: logs/multiruns/${core.name}/${now:%Y-%m-%d_%H-%M-%S}
+    subdir: ${multirun_subdir_beautify:${hydra.job.override_dirname}}
+  job:
+    config:
+      override_dirname:
+        exclude_keys:
+          - trainer.devices
+    chdir: true

quadra/configs/inference/default.yaml

@@ -0,0 +1,26 @@
+onnx:
+  session_options:
+    inter_op_num_threads: 8
+    intra_op_num_threads: 8
+    graph_optimization_level:
+      _target_: onnxruntime.GraphOptimizationLevel
+      value: 99 # ORT_ENABLE_ALL
+    enable_mem_pattern: true
+    enable_cpu_mem_arena: true
+    enable_profiling: false
+    enable_mem_reuse: true
+    execution_mode:
+      _target_: onnxruntime.ExecutionMode
+      value: 0 # ORT_SEQUENTIAL
+    execution_order:
+      _target_: onnxruntime.ExecutionOrder
+      value: 0 # DEFAULT
+    log_severity_level: 2
+    log_verbosity_level: 0
+    logid: ""
+    optimized_model_filepath: ""
+    use_deterministic_compute: false
+    profile_file_prefix: onnxruntime_profile_
+
+pytorch:
+torchscript:

quadra/configs/logger/comet.yaml

@@ -0,0 +1,10 @@
+comet:
+  _target_: pytorch_lightning.loggers.CometLogger
+  api_key: ${oc.env:COMET_API_KEY}
+  save_dir: null
+  project_name: ${core.name}
+  rest_api_key: null
+  experiment_name: ${core.name}
+  experiment_key: null
+  offline: false
+  prefix: ""

quadra/configs/logger/csv.yaml

@@ -0,0 +1,5 @@
+csv:
+  _target_: pytorch_lightning.loggers.csv_logs.CSVLogger
+  save_dir: "."
+  name: "csv/"
+  prefix: ""

quadra/configs/logger/mlflow.yaml

@@ -0,0 +1,12 @@
+# https://mlflow.org
+
+mlflow:
+  _target_: pytorch_lightning.loggers.mlflow.MLFlowLogger
+  experiment_name: ${core.name}
+  tracking_uri: ${oc.env:MLFLOW_TRACKING_URI}
+  tags: null
+  save_dir: null
+  prefix: ""
+  artifact_location: null
+  run_id: null
+  log_model: false

quadra/configs/logger/tensorboard.yaml

@@ -0,0 +1,8 @@
+tensorboard:
+  _target_: pytorch_lightning.loggers.tensorboard.TensorBoardLogger
+  save_dir: "tensorboard/"
+  name: ${core.name}
+  version: null
+  log_graph: False
+  default_hp_metric: True
+  prefix: ""

quadra/configs/loss/asl.yaml

@@ -0,0 +1,7 @@
+_target_: quadra.losses.classification.AsymmetricLoss
+gamma_neg: 4
+gamma_pos: 0
+m: 0.05
+eps: 1e-8
+disable_torch_grad_focal_loss: false
+apply_sigmoid: true

quadra/configs/loss/barlow.yaml

@@ -0,0 +1,2 @@
+_target_: quadra.losses.ssl.BarlowTwinsLoss
+lambd: 5e-3

quadra/configs/loss/bce.yaml

@@ -0,0 +1 @@
+_target_: torch.nn.BCEWithLogitsLoss

quadra/configs/loss/byol.yaml

@@ -0,0 +1 @@
+_target_: quadra.losses.ssl.BYOLRegressionLoss

quadra/configs/loss/cross_entropy.yaml

@@ -0,0 +1 @@
+_target_: torch.nn.CrossEntropyLoss

quadra/configs/loss/dino.yaml

@@ -0,0 +1,8 @@
+_target_: quadra.losses.ssl.DinoDistillationLoss
+output_dim: ${model.student_projection_mlp.output_dim}
+max_epochs: ${trainer.max_epochs}
+warmup_teacher_temp: 0.04
+teacher_temp: 0.07
+warmup_teacher_temp_epochs: 30
+student_temp: 0.1
+center_momentum: 0.9

quadra/configs/loss/simclr.yaml

@@ -0,0 +1,2 @@
+_target_: quadra.losses.ssl.SimCLRLoss
+temperature: 0.1

quadra/configs/loss/simsiam.yaml

@@ -0,0 +1 @@
+_target_: quadra.losses.ssl.SimSIAMLoss

quadra/configs/loss/smp_ce.yaml

@@ -0,0 +1,3 @@
+_target_: segmentation_models_pytorch.losses.SoftCrossEntropyLoss
+dim: 1
+smooth_factor: 0.01

quadra/configs/loss/smp_dice.yaml

@@ -0,0 +1,2 @@
+_target_: segmentation_models_pytorch.losses.DiceLoss
+mode: binary

quadra/configs/loss/smp_dice_multiclass.yaml

@@ -0,0 +1,2 @@
+_target_: segmentation_models_pytorch.losses.DiceLoss
+mode: multiclass

quadra/configs/loss/smp_mcc.yaml

@@ -0,0 +1,2 @@
+_target_: segmentation_models_pytorch.losses.MCCLoss
+eps: 1e-05

quadra/configs/loss/vicreg.yaml

@@ -0,0 +1,5 @@
+_target_: quadra.losses.ssl.VICRegLoss
+lambd: 25
+mu: 25
+nu: 1
+gamma: 1

quadra/configs/model/anomalib/cfa.yaml

@@ -0,0 +1,35 @@
+dataset:
+  task: segmentation
+
+model:
+  input_size: [224, 224]
+  backbone: resnet18.tv_in1k
+  gamma_c: 1
+  gamma_d: 1
+  num_nearest_neighbors: 3
+  num_hard_negative_features: 3
+  radius: 1e-5
+  lr: 1e-3
+  weight_decay: 5e-4
+  amsgrad: true
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null
+
+# PL Trainer Args. Don't add extra parameter here.
+trainer:
+  gradient_clip_val: 0
+  gradient_clip_algorithm: norm
+  devices: [0]
+  check_val_every_n_epoch: 1 # Don't validate before extracting features.
+  max_epochs: 30
+  val_check_interval: 1.0 # Don't validate before extracting features.

quadra/configs/model/anomalib/cflow.yaml

@@ -0,0 +1,30 @@
+dataset:
+  task: segmentation
+
+model:
+  name: cflow
+  backbone: resnet18.tv_in1k
+  input_size: [256, 256]
+  layers:
+    - layer2
+    - layer3
+    - layer4
+  decoder: freia-cflow
+  condition_vector: 128
+  coupling_blocks: 8
+  permute_soft: false
+  clamp_alpha: 1.9
+  lr: 0.0001
+  fiber_batch_size: 64
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/anomalib/csflow.yaml

@@ -0,0 +1,34 @@
+dataset:
+  task: segmentation
+  tiling:
+    apply: false
+    tile_size: null
+    stride: null
+    remove_border_count: 0
+    use_random_tiling: False
+    random_tile_count: 16
+
+model:
+  name: csflow
+  clamp: 3
+  input_size: [256, 256] # 768 is the dimensions used in the official implementation
+  cross_conv_hidden_channels: 1024
+  eps: 1e-04 # Adam epsilon
+  layers:
+    - 6.8
+  lr: 2e-4
+  n_coupling_blocks: 4
+  n_scales: 3
+  weight_decay: 1e-5 # Adam weight decay
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive #options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/anomalib/dfm.yaml

@@ -0,0 +1,19 @@
+dataset:
+  task: classification
+
+model:
+  backbone: wide_resnet50_2.tv_in1k
+  pca_level: 0.97
+  score_type: fre # nll: for Gaussian modeling, fre: pca feature reconstruction error
+  threshold:
+    image_default: 0
+    pixel_default: 0
+    adaptive: true
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC

quadra/configs/model/anomalib/draem.yaml

@@ -0,0 +1,29 @@
+dataset:
+  task: segmentation
+  tiling:
+    apply: false
+    tile_size: null
+    stride: null
+    remove_border_count: 0
+    use_random_tiling: False
+    random_tile_count: 16
+
+model:
+  name: draem
+  input_size: [256, 256]
+  anomaly_source_path: null # optional, e.g. ./datasets/dtd
+  lr: 0.0001
+  enable_sspcab: false
+  sspcab_lambda: 0.1
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/anomalib/efficient_ad.yaml

@@ -0,0 +1,31 @@
+dataset:
+  task: segmentation
+
+model:
+  name: efficientad
+  teacher_out_channels: 384
+  model_size: small # options: [small, medium]
+  lr: 0.0001
+  input_size: [256, 256]
+  weight_decay: 0.00001
+  padding: false
+  pad_maps: true # relevant for "padding: false", see EfficientAd in lightning_model.py
+  pre_padding: true # set True to avoid anomaly maps' dead frame issue
+  # generic params
+  normalization_method: min_max # options: [null, min_max, cdf]
+  train_batch_size: 1 # ${datamodule.train_batch_size}
+  pretrained_models_dir: ${oc.env:HOME}/.quadra/models/efficient_ad
+  imagenette_dir: ${oc.env:HOME}/.quadra/datasets/imagenette_efficientad
+  pretrained_teacher_type: nelson
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/anomalib/fastflow.yaml

@@ -0,0 +1,32 @@
+dataset:
+  task: segmentation
+  tiling:
+    apply: false
+    tile_size: null
+    stride: null
+    remove_border_count: 0
+    use_random_tiling: False
+    random_tile_count: 16
+
+model:
+  name: fastflow
+  input_size: [224, 224]
+  backbone: resnet18.tv_in1k # options: [resnet18, wide_resnet50_2, cait_m48_448, deit_base_distilled_patch16_384]
+  pre_trained: true
+  flow_steps: 8 # options: [8, 8, 20, 20] - for each supported backbone
+  hidden_ratio: 1.0 # options: [1.0, 1.0, 0.16, 0.16] - for each supported backbone
+  conv3x3_only: True # options: [True, False, False, False] - for each supported backbone
+  lr: 0.001
+  weight_decay: 0.00001
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/anomalib/padim.yaml

@@ -0,0 +1,32 @@
+dataset:
+  task: segmentation
+  tiling:
+    apply: false
+    tile_size: null
+    stride: null
+    remove_border_count: 0
+    use_random_tiling: False
+    random_tile_count: 16
+
+model:
+  name: padim
+  input_size: [224, 224]
+  backbone: resnet18.tv_in1k
+  layers:
+    - layer1
+    - layer2
+    - layer3
+  pre_trained: true
+  n_features: null
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/anomalib/patchcore.yaml

@@ -0,0 +1,36 @@
+dataset:
+  task: segmentation
+  tiling:
+    apply: false
+    tile_size: null
+    stride: null
+    remove_border_count: 0
+    use_random_tiling: False
+    random_tile_count: 16
+
+model:
+  name: patchcore
+  backbone: resnet18.tv_in1k
+  layers:
+    - layer2
+    - layer3
+  input_size: [224, 224]
+  coreset_sampling_ratio: 0.1
+  num_neighbors: 9
+  pre_trained: true
+  pretrained_weights:
+  compress_memory_bank: false # Whether to apply dimensionality reduction to the memory bank
+  coreset_sampler: anomalib # Can be this or amazon
+  score_computation: anomalib # Can be this or amazon
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive # options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null

quadra/configs/model/barlow.yaml

@@ -0,0 +1,16 @@
+model: ${backbone.model}
+projection_mlp_mult: 4
+projection_mlp:
+  _target_: quadra.modules.ssl.common.BarlowTwinsProjectionHead
+  input_dim: ${backbone.metadata.output_dim}
+  output_dim: ${backbone.metadata.output_dim}
+  hidden_dim: ${backbone.metadata.output_dim}
+module:
+  _target_: quadra.modules.ssl.BarlowTwins
+  classifier: ${model.classifier}
+  criterion: ${loss}
+  lr_scheduler_interval: step
+classifier:
+  _target_: sklearn.neighbors.KNeighborsClassifier
+  n_neighbors: 20
+  n_jobs: 5

quadra/configs/model/byol.yaml

@@ -0,0 +1,25 @@
+projection_mlp:
+  _target_: quadra.modules.ssl.common.BYOLProjectionHead
+  input_dim: ${backbone.metadata.output_dim}
+  output_dim: 256
+  hidden_dim: 4096
+
+prediction_mlp:
+  _target_: quadra.modules.ssl.common.BYOLPredictionHead
+  input_dim: ${model.projection_mlp.output_dim}
+  output_dim: ${model.projection_mlp.output_dim}
+  hidden_dim: ${model.projection_mlp.hidden_dim}
+
+student: ${backbone.model}
+
+module:
+  _target_: quadra.modules.ssl.BYOL
+  classifier: ${model.classifier}
+  criterion: ${loss}
+  teacher_momentum: 0.9996
+  lr_scheduler_interval: step
+
+classifier:
+  _target_: sklearn.neighbors.KNeighborsClassifier
+  n_neighbors: 20
+  n_jobs: 5