| 1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192 |
- Global:
- use_gpu: True
- epoch_num: 8
- log_smooth_window: 20
- print_batch_step: 5
- save_model_dir: ./output/rec/pren_new
- save_epoch_step: 3
- # evaluation is run every 2000 iterations after the 4000th iteration
- eval_batch_step: [4000, 2000]
- cal_metric_during_train: True
- pretrained_model:
- checkpoints:
- save_inference_dir:
- use_visualdl: False
- infer_img: doc/imgs_words/ch/word_1.jpg
- # for data or label process
- character_dict_path:
- max_text_length: &max_text_length 25
- infer_mode: False
- use_space_char: False
- save_res_path: ./output/rec/predicts_pren.txt
- Optimizer:
- name: Adadelta
- lr:
- name: Piecewise
- decay_epochs: [2, 5, 7]
- values: [0.5, 0.1, 0.01, 0.001]
- Architecture:
- model_type: rec
- algorithm: PREN
- in_channels: 3
- Backbone:
- name: EfficientNetb3_PREN
- Neck:
- name: PRENFPN
- n_r: 5
- d_model: 384
- max_len: *max_text_length
- dropout: 0.1
- Head:
- name: PRENHead
- Loss:
- name: PRENLoss
- PostProcess:
- name: PRENLabelDecode
- Metric:
- name: RecMetric
- main_indicator: acc
- Train:
- dataset:
- name: LMDBDataSet
- data_dir: ./train_data/data_lmdb_release/training/
- transforms:
- - DecodeImage:
- img_mode: BGR
- channel_first: False
- - PRENLabelEncode:
- - RecAug:
- - PRENResizeImg:
- image_shape: [64, 256] # h,w
- - KeepKeys:
- keep_keys: ['image', 'label']
- loader:
- shuffle: True
- batch_size_per_card: 128
- drop_last: True
- num_workers: 8
- Eval:
- dataset:
- name: LMDBDataSet
- data_dir: ./train_data/data_lmdb_release/validation/
- transforms:
- - DecodeImage:
- img_mode: BGR
- channel_first: False
- - PRENLabelEncode:
- - PRENResizeImg:
- image_shape: [64, 256] # h,w
- - KeepKeys:
- keep_keys: ['image', 'label']
- loader:
- shuffle: False
- drop_last: False
- batch_size_per_card: 64
- num_workers: 8
|
