Merge remote-tracking branch 'origin/master' into main

nndet/conf/config.yaml

+defaults:
+    - train: v001
+    - prep: process
+
+exp:
+  tag: ""
+  fold: 0
+  id: ${module}_${plan}${exp.tag}
+
+train:
+  mode: "overwrite" # can be either `overwrite` or `resume`
+  val_test: True # run testing on validation data to get final training results
+
+host:
+  # DO NOT CHANGE THESE
+  parent_data: ${oc.env:det_data}
+  parent_results: ${oc.env:det_models}
+
+  # base is the folder where the raw data is stored
+  data_dir: ${host.parent_data}/${task}
+  prep_dir: ${host.parent_data}/${task}
+  network_training_output_dir: ${host.parent_results} # kept for historical reasons
+
+  # create intermediate folder for preprocessing steps
+  raw_output_dir: ${host.prep_dir}/raw
+  splitted_4d_output_dir: ${host.prep_dir}/raw_splitted
+  cropped_output_dir: ${host.prep_dir}/raw_cropped
+  # preprocessing_output_dir is where the preprocessed data is stored. If you run a training I very strongly recommend
+  # this is a SSD!
+  preprocessed_output_dir: ${host.parent_data}/${task}/preprocessed
+
+  # network training
+  plan_path: ${host.preprocessed_output_dir}/${plan}.pkl

nndet/conf/prep/nothing.yaml

+# set this to 1 if you want to override cropped data
+overwrite: False
+
+crop: False
+analyze: False
+plan: False
+process: False

nndet/conf/prep/plan.yaml

+# set this to 1 if you want to override cropped data
+overwrite: False
+
+crop: False
+analyze: False
+plan: True
+process: False

nndet/conf/prep/process.yaml

+# set this to 1 if you want to override cropped data
+overwrite: False
+
+crop: True
+analyze: True
+plan: True
+process: True

nndet/conf/train/augmentation/base_more.yaml

+name: "base_more"
+transforms: "BaseMoreAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: False
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.2
+do_scaling: True
+scale_range: [0.7, 1.4]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: True
+rotation_x: [-30, 30]
+rotation_y: [-30, 30]
+rotation_z: [-30, 30]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/base_more_sol.yaml

+name: "base_more_sol"
+transforms: "BaseMoreAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: False
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.2
+do_scaling: True
+scale_range: [0.7, 1.]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: True
+rotation_x: [-30, 30]
+rotation_y: [-30, 30]
+rotation_z: [-30, 30]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/base_more_sou.yaml

+name: "base_more_sou"
+transforms: "BaseMoreAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: False
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.2
+do_scaling: True
+scale_range: [1., 1.4]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: True
+rotation_x: [-30, 30]
+rotation_y: [-30, 30]
+rotation_z: [-30, 30]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/base_mp.yaml

+name: "base_mp" # identifier of this specific set of params for config
+transforms: "BaseMoreAug" # name of the augmentation modue to use
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: False
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.3
+do_scaling: True
+scale_range: [0.85, 1.25]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.3
+do_rotation: True
+rotation_x: [-15, 15]
+rotation_y: [-15, 15]
+rotation_z: [-15, 15]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/default.yaml

+name: "default"
+transforms: "DefaultAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: True
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.2
+do_scaling: True
+scale_range: [0.85, 1.25]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: True
+rotation_x: [-15, 15]
+rotation_y: [-15, 15]
+rotation_z: [-15, 15]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/insane.yaml

+name: "insane"
+transforms: "InsaneAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: True
+elastic_deform_alpha: [0., 1300.]
+elastic_deform_sigma: [9., 15.]
+
+p_scale: 0.2
+do_scaling: True
+scale_range: [0.65, 1.6]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: True
+rotation_x: [-30, 30]
+rotation_y: [-30, 30]
+rotation_z: [-30, 30]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.6, 2]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/mirror_only.yaml

+name: "mirror_only"
+transforms: "DefaultAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: False
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.2
+do_scaling: False
+scale_range: [0.85, 1.25]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: False
+rotation_x: [-15, 15]
+rotation_y: [-15, 15]
+rotation_z: [-15, 15]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: False
+gamma_retain_stats: False
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/more.yaml

+name: "more"
+transforms: "MoreAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+p_eldef: 0.2
+do_elastic: False
+elastic_deform_alpha: [0., 900.]
+elastic_deform_sigma: [9., 13.]
+
+p_scale: 0.2
+do_scaling: True
+scale_range: [0.7, 1.4]
+independent_scale_factor_for_each_axis: False
+
+p_rot: 0.2
+do_rotation: True
+rotation_x: [-30, 30]
+rotation_y: [-30, 30]
+rotation_z: [-30, 30]
+
+order_data: 3
+border_mode_data: "constant"
+border_cval_data: 0
+order_seg: 0
+border_cval_seg: -1
+border_mode_seg: "constant"
+
+random_crop: False
+random_crop_dist_to_border:
+
+p_gamma: 0.3
+do_gamma: True
+gamma_retain_stats: True
+gamma_range: [0.7, 1.5]
+
+do_mirror: True
+mirror_axes: [0, 1, 2]
+
+do_additive_brightness: False
+additive_brightness_p_per_sample: 0.15
+additive_brightness_p_per_channel: 0.5
+additive_brightness_mu: 0.0
+additive_brightness_sigma: 0.1
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 200.]
+  elastic_deform_sigma: [9., 13.]
+  rotation_x: [-180, 180]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: [0, 1]

nndet/conf/train/augmentation/noaug.yaml

+name: "noaug"
+transforms: "NoAug"
+transforms_kwargs: {}
+
+selected_data_channels:
+selected_seg_channels:
+
+rotation_x: [-0, 0]
+rotation_y: [-0, 0]
+rotation_z: [-0, 0]
+scale_range: [1., 1.]
+
+2d_overwrites:
+  elastic_deform_alpha: [0., 0.]
+  elastic_deform_sigma: [0., 0.]
+  rotation_x: [0, 0]
+  rotation_y: [0, 0]
+  rotation_z: [0, 0]
+  dummy_2D: False
+  mirror_axes: []

nndet/conf/train/deprecated/h0c03.yaml

+# @package __global__
+defaults:
+  - augmentation: base
+
+model: "RetinaUNetC008"
+trainer: "DetectionTrainerPolyLR"
+predictor: "BoxPredictorSelective"
+plan: D3C002_3d
+
+planners:
+  2d: [D2C002]
+  3d: [D2C002, D3C002] # [D3C002LR15, D3C002LR20] [D3C002NR, D3C002RibFrac] [D2C002, D3C002]
+
+augment_cfg:
+  oversample_foreground_percent: 0.5 # ratio of fg and bg in batches
+  augmentation: ${augmentation}
+  dataloader: "DataLoader{}DFast"
+  dataloader_kwargs: {}
+
+trainer_cfg:
+  # Per default training is deterministic, non-deterministic allows
+  # cudnn.benchmark which can give up to 20% performance. Set this to false
+  # to perform non-deterministic training
+  deterministic: True
+  fp16: True # enable fp16 training. Makes sense for supported hardware only!
+  eval_score_key: "mAP_IoU_0.10_0.50_0.05_MaxDet_100" # metric to optimize
+
+  num_batches_per_epoch: 2500 # number of train batches per epoch
+  num_val_batches_per_epoch: 100 # number of val batches per epoch
+  max_num_epochs: 50 # max number of epochs
+  overwrites: {}
+
+  initial_lr: 3.e-4 # initial learning rate to start with
+  weight_decay: 3.e-5 # weight decay for optimizer
+
+  warmup: 4000 # number of iterations with warmup
+  warmup_lr: 1.e-6 # learning rate to start warmup from
+
+model_cfg:
+  matching:
+    # IoU Matcher Parameters
+    fg_iou_thresh: 0.4 # IoU threshold for anchors to be matched positive
+    bg_iou_thresh: 0.3 # IoU threshold for anchors to be matched negative
+    # If ground truth has no matched anchors, use the best anchor which was found
+    allow_low_quality_matches: True
+
+    # ATSS matching
+    num_candidates: 4
+    center_in_gt: False
+
+  hnm: # parameters for hard negative mining
+    batch_size_per_image: 32 # number of anchors sampled per image
+    positive_fraction: 0.33 # defines ratio between positive and negative anchors
+    # hard negatives are sampled from a pool of size:
+    # batch_size_per_image * (1 - positive_fraction) * pool_size
+    pool_size: 20
+    min_neg: 1 # minimum number of negative anchors sampled per image
+
+  plan_arch_overwrites: {} # overwrite arguments of architecture
+  plan_anchors_overwrites: {} # overwrite arguments of anchors

nndet/conf/train/deprecated/h0c03_aniso.yaml

+model: "RetinaUNetC008"
+trainer: "DetectionTrainerPolyLR"
+predictor: "BoxPredictorSelective"
+plan: D3C002_3d
+
+planners:
+  2d: [D2C002]
+  3d: [D3C002, D2C002]
+
+augment_cfg:
+  oversample_foreground_percent: 0.5 # ratio of fg and bg in batches
+  augmentation: ${augmentation}
+  dataloader: "DataLoader{}DFast"
+  dataloader_kwargs: {}
+
+trainer_cfg:
+  # Per default training is deterministic, non-deterministic allows
+  # cudnn.benchmark which can give up to 20% performance. Set this to false
+  # to perform non-deterministic training
+  deterministic: True
+  fp16: True # enable fp16 training. Makes sense for supported hardware only!
+  eval_score_key: "mAP_IoU_0.10_0.50_0.05_MaxDet_100" # metric to optimize
+
+  num_batches_per_epoch: 2500 # number of train batches per epoch
+  num_val_batches_per_epoch: 100 # number of val batches per epoch
+  max_num_epochs: 50 # max number of epochs # CHANGE TEMP
+  overwrites: {}
+
+  initial_lr: 3.e-4 # initial learning rate to start with
+  weight_decay: 3.e-5 # weight decay for optimizer
+
+  warmup: 4000 # number of iterations with warmup
+  warmup_lr: 1.e-6 # learning rate to start warmup from
+
+model_cfg:
+  matching:
+    # IoU Matcher Parameters
+    fg_iou_thresh: 0.4 # IoU threshold for anchors to be matched positive
+    bg_iou_thresh: 0.3 # IoU threshold for anchors to be matched negative
+    # If ground truth has no matched anchors, use the best anchor which was found
+    allow_low_quality_matches: True
+
+    # ATSS matching
+    num_candidates: 4
+    center_in_gt: False
+
+  hnm: # parameters for hard negative mining
+    batch_size_per_image: 32 # number of anchors sampled per image
+    positive_fraction: 0.33 # defines ratio between positive and negative anchors
+    # hard negatives are sampled from a pool of size:
+    # batch_size_per_image * (1 - positive_fraction) * pool_size
+    pool_size: 20
+    min_neg: 1 # minimum number of negative anchors sampled per image
+
+  plan_arch_overwrites: # overwrite arguments of architecture
+      strides: [[1, 2, 2], [1, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]]
+      conv_kernels: [[1, 3, 3], [1, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [1, 3, 3]]
+      decoder_levels: [2, 3, 4, 5]
+
+  plan_anchors_overwrites: # overwrite arguments of anchors
+    width: [[2.0, 3.0, 4.0], [4.0, 6.0, 8.0], [8.0, 12.0, 16.0], [8.0, 12.0, 16.0]]
+    height: [[3.0, 4.0, 5.0], [6.0, 8.0, 10.0], [12.0, 16.0, 20.0], [24.0, 32.0, 40.0]]
+    depth: [[3.0, 4.0, 5.0], [6.0, 8.0, 10.0], [12.0, 16.0, 20.0], [24.0, 32.0, 40.0]]

nndet/conf/train/deprecated/h0c04.yaml

+# @package __global__
+defaults:
+  - augmentation: base
+
+model: "RetinaUNetC009"
+trainer: "DetectionTrainerPolyLR"
+predictor: "BoxPredictorSelective"
+plan: D3C002_3d
+
+planners:
+  2d: [D2C002] # D2C002 D2C002LR20
+  3d: [D2C002, D3C002] # [D3C002LR15, D3C002LR20] [D3C002NR, D3C002RibFrac] [D2C002, D3C002]
+
+augment_cfg:
+  oversample_foreground_percent: 0.5 # ratio of fg and bg in batches
+  augmentation: ${augmentation}
+  dataloader: "DataLoader{}DFast"
+  dataloader_kwargs: {}
+
+trainer_cfg:
+  # Per default training is deterministic, non-deterministic allows
+  # cudnn.benchmark which can give up to 20% performance. Set this to false
+  # to perform non-deterministic training
+  deterministic: True
+  fp16: True # enable fp16 training. Makes sense for supported hardware only!
+  eval_score_key: "mAP_IoU_0.10_0.50_0.05_MaxDet_100" # metric to optimize
+
+  num_batches_per_epoch: 2500 # number of train batches per epoch
+  num_val_batches_per_epoch: 100 # number of val batches per epoch
+  max_num_epochs: 50 # max number of epochs
+  overwrites: {}
+
+  initial_lr: 3.e-4 # initial learning rate to start with
+  weight_decay: 3.e-5 # weight decay for optimizer
+
+  warmup: 4000 # number of iterations with warmup
+  warmup_lr: 1.e-6 # learning rate to start warmup from
+
+model_cfg:
+  matching:
+    # IoU Matcher Parameters
+    fg_iou_thresh: 0.4 # IoU threshold for anchors to be matched positive
+    bg_iou_thresh: 0.3 # IoU threshold for anchors to be matched negative
+    # If ground truth has no matched anchors, use the best anchor which was found
+    allow_low_quality_matches: True
+
+    # ATSS matching
+    num_candidates: 4
+    center_in_gt: False
+
+  hnm: # parameters for hard negative mining
+    batch_size_per_image: 32 # number of anchors sampled per image
+    positive_fraction: 0.33 # defines ratio between positive and negative anchors
+    # hard negatives are sampled from a pool of size:
+    # batch_size_per_image * (1 - positive_fraction) * pool_size
+    pool_size: 20
+    min_neg: 1 # minimum number of negative anchors sampled per image
+
+  plan_arch_overwrites: {} # overwrite arguments of architecture
+  plan_anchors_overwrites: {} # overwrite arguments of anchors

nndet/conf/train/deprecated/h0c04SGD.yaml

+# @package __global__
+defaults:
+  - augmentation: base_more
+
+model: "RetinaUNetC009"
+trainer: "DetectionTrainerPolyLR_SGD090"
+predictor: "BoxPredictorSelective"
+plan: D3C002_3d
+
+planners:
+  2d: [D2C002] # D2C002 D2C002LR20
+  3d: [D2C002, D3C002] # [D3C002LR15, D3C002LR20] [D3C002NR, D3C002RibFrac] [D2C002, D3C002]
+
+augment_cfg:
+  oversample_foreground_percent: 0.5 # ratio of fg and bg in batches
+  augmentation: ${augmentation}
+  dataloader: "DataLoader{}DFast"
+  dataloader_kwargs: {}
+
+trainer_cfg:
+  # Per default training is deterministic, non-deterministic allows
+  # cudnn.benchmark which can give up to 20% performance. Set this to false
+  # to perform non-deterministic training
+  deterministic: True
+  fp16: True # enable fp16 training. Makes sense for supported hardware only!
+  eval_score_key: "mAP_IoU_0.10_0.50_0.05_MaxDet_100" # metric to optimize
+
+  num_batches_per_epoch: 2500 # number of train batches per epoch
+  num_val_batches_per_epoch: 100 # number of val batches per epoch
+  max_num_epochs: 50 # max number of epochs
+  overwrites: {}
+
+  initial_lr: 0.01 # initial learning rate to start with
+  weight_decay: 3.e-5 # weight decay for optimizer
+
+  warmup: 4000 # number of iterations with warmup
+  warmup_lr: 1.e-6 # learning rate to start warmup from
+
+model_cfg:
+  matching:
+    # IoU Matcher Parameters
+    fg_iou_thresh: 0.4 # IoU threshold for anchors to be matched positive
+    bg_iou_thresh: 0.3 # IoU threshold for anchors to be matched negative
+    # If ground truth has no matched anchors, use the best anchor which was found
+    allow_low_quality_matches: True
+
+    # ATSS matching
+    num_candidates: 4
+    center_in_gt: False
+
+  hnm: # parameters for hard negative mining
+    batch_size_per_image: 32 # number of anchors sampled per image
+    positive_fraction: 0.33 # defines ratio between positive and negative anchors
+    # hard negatives are sampled from a pool of size:
+    # batch_size_per_image * (1 - positive_fraction) * pool_size
+    pool_size: 20
+    min_neg: 1 # minimum number of negative anchors sampled per image
+
+  plan_arch_overwrites: {} # overwrite arguments of architecture
+  plan_anchors_overwrites: {} # overwrite arguments of anchors