config.toml

[PSO]
random_radius = 5
population_number = 5
iter_number = 30
padding_step = 2

[DATAEXTRACT]
data_origin_path = 'data/origin'
data_square_path = 'data/square'
data_circle_path = 'data/circle'
data_split_train_path = 'data/train'
data_split_valid_path = 'data/valid'
data_split_origin_path = 'data/square'
#  train_ration可以小很多，如果是静态构建的数据集训练的话
train_ratio = 0.8
clear_origin = false
# 224 224 before
img_size = [128, 128]

[MODEL]
#  resnet18, resnet34, resnet50, resnet101, resnet152
model_type = 'resnet18'
feature_dim = 128
pretrained = true

[TRAIN]
save_weights = "runs"
train_path = 'data/train'
valid_path = 'data/valid'
device = 'cuda:0'
batch_size = 16
back_true = 0.5
works = 1
# 支持的损失函数：CosineSimilarityLoss，ClassFiyOneLoss，
#ClassFiyTwoLoss，CosineMarginOneLoss，CosineMarginTwoLoss,PalmCombinedLoss(当前项目默认的损失函数)
loss = 'PalmCombinedLoss'
similarity_threshold = 0.75
loss_margin = 0.0
shuffle = true
lr = 0.001
epochs = 200
val_interval = 10
log_interval = 200
save_epoch = 10

[DETECT]
device = 'cuda:0'