config/PRAKA.yaml

dataset: &dataset cifar100
init_cls_num: &init_cls_num 50 # 50  50
inc_cls_num: &inc_cls_num 10 # 5   10
total_cls_num: &total_cls_num 100 
task_num: &task_num 6 # 11  6
image_size: &image_size 32

image_size: *image_size
# data
init_cls_num: *init_cls_num  
inc_cls_num: *inc_cls_num   
task_num: *task_num  
batch_size: 128 # 128
epoch: 100 # 100

val_per_epoch: 100
seed: 2

testing_times: 10 # 10

train_trfms:
  - RandomCrop: 
      size: [*image_size, *image_size]
      padding: 4
  - RandomHorizontalFlip: 
      p: 0.5
  - ColorJitter: 
      brightness: 0.24705882352941178
  - ToTensor: {}
  - Normalize:
      mean: [0.5071, 0.4866, 0.4409] # don't change
      std: [0.2675, 0.2565, 0.2761] # don't change
      #mean: [0.5071, 0.4867, 0.4408]
      #std: [0.2675, 0.2565, 0.2761]


test_trfms: 
  - ToTensor: {}
  - Normalize:
      mean: [0.5071, 0.4866, 0.4409] # don't change
      std: [0.2675, 0.2565, 0.2761] # don't change
      #mean: [0.5071, 0.4867, 0.4408]
      #std: [0.2675, 0.2565, 0.2761]

optimizer:
  name: Adam
  kwargs:
    lr: 0.001
    #betas: [0.9, 0.999]
    weight_decay: 2e-4
    #eps: 1e-8

lr_scheduler:
  name: CosineAnnealingLR
  kwargs:
    T_max: 32

backbone:
  name: resnet18_cbam
  kwargs:
    num_classes: *total_cls_num
    args:
      dataset: *dataset

classifier:
  name: PRAKA
  kwargs:
    num_class: *total_cls_num
    init_cls_num: *init_cls_num
    inc_cls_num: *inc_cls_num
    feat_dim: 512

    log_root: log
    total_nc: *total_cls_num
    protoAug_weight: 15.0
    kd_weight: 15.0
    temp: 0.1