v2_3.yml

NF: 3
VCL: 8
apply_softmax: true
batch_size: 8
class_weights: (0.3, 0.7)
data_mean: [.5, .5, .5]
data_path: ./data/dota
data_std: [.5, .5, .5]
dataset: Dota
dropout: 0.3
image_shape: [720, 1280]
input_shape: [240, 320]
lr: 0.0001
pretrained: 'pretrained/swin_base_patch244_window1677_sthv2.pth'
rnn_cell_num: 3
rnn_state_size: 1024
transformer_type: 'swin_base_patch244_window1677_sthv2'
vertical_flip_prob: 0.5