config_conic.sys

log = "log/"

[Debug]
debug = "False"                                 # Debug Mode - True or False: Radically reduces the dataset size for faster training
debugDilution = 200                            # Debug Dilution - 1/nth of the dataset will be used for training
wandb = "True"                                 # Wandb Logging

[Stain Normalization]
normalization = "reinhard"                       # Normalization Avl: macenko, reinhard, None
targetImagePath = "Dataset/MonuSegData/Training/TissueImages/TCGA-A7-A13F-01Z-00-DX1.png" # Target Image for Normalization


[Image Augmentation - CoNIC]
imagePath = "Dataset/Otherdata/CoNICChallengeTrain/images.npy"
labelPath = "Dataset/Otherdata/CoNICChallengeTrain/labels.npy"
augmentPerImage = 10                           # Number of augmentation to perform per image that was generated using sliding down
finalTileHeight = 256                           # Final Height of augmenated image
finalTileWidth = 256                            # Final Width of augmenated image


[Train-Val split]
splitRatio = 0.7                                # 0.8 -> 80% training 20% validation

[Training]
trainDataset = "Dataset/Otherdata/prepared_CoNIC/trainNormal/"                    # Images to be during Training
valDataset = "Dataset/Otherdata/prepared_CoNIC/valNormal/"                        # Images to used during Validation
testDataset = "Dataset/Otherdata/prepared_CoNIC/testNormal/"                      # Images to used during Testing

resumeModel = "model/best_model.pth"            # Model to used during re-training

torchsummary = "False"                          # Torch Summary
sampleImages = "False"                           # Sample Images during Training
dinoModelType = "giga"                         # DINO Model Type - small, large, giga
reUseFeatures = "True"                         # Use existing features - Should re-train features if dataset is changed
batchVisualization = "True"                   # Sampling Visualization
trainingPhase = "high-density"                  # Training Phase - high-density, low-density

[Class Config]
class1 = "0, 0, 0"                          #black
class2 = "255, 255, 255"                         #white

[Model]
model_type = "UNet"                           # Model Avl: UNet ,UNet_3Plus, EluNet, UNet_3PlusShort

[Parameters]
input_img_type = "rgb"                          # Input Image Type: rgb, gray
kernel_size = 3
use_maxblurpool = "False"                       # Use MaxBlurPool
epochs = 50
batch_size = 16
learning_rate = 0.0000001                         # can be set to "auto" for finding the best learning rate or set to 0.00001
lr_decay = "False"                             # Learning Rate Decay  - True or False
num_classes = 2
weight_path = "None"  # Path to the model else None
activation = "relu"                             # Activations available: relu, GLU
resume = "False"                                # Resume Training
resume_epoch = 0                                # Resume Training from epoch
channel = 32                                    # Default: 16 Can be set to 64 or other
attention = "False"                             # Model 1 only
loss = "bce"                           # Losses available: bce, jaccard, focal, pwcel, dice, weighteddice, unet3+loss, improvedLoss, ClassRatioLoss, RBAF, focalDiceLoss, wassersteinLoss, focalDiceHDLoss
dropout = 0.3                                   # Dropout rate
dropoutLOC = "std"                              # Dropout Location: std, after (after 20 iterations)
dilation = 1                                    # For Dilated convolution.
l1_regularization = "False"                     # L1 Regularization

[experiment features]
multiScaleAttention = "False"                   # Spatial, Channel and Edge Attention used in tandem

eigen_decomposition = "False"                  # Eigen Decomposition
top_k_features = 0                        # Top K Features - 0- False, n-Top N values

dropBlock = "True"                             # DropBlock, automatically sets dropout to 0 and proceeds with dropblock
dropBlockProb = 0.8                             # DropBlock Probability
dropBlockSize = 5                            # DropBlock Size

lookahead = "False"                             # Lookahead
lookahead_k = 5                                 # Lookahead k
lookahead_alpha = 0.8                           # Lookahead alpha

guidedFilter = "True"                         # Guilded Filter