DETR:
  backbone: PPHGNetV2
  detr_head: DocLayoutV3Head
  neck: MaskHybridEncoder
  post_process: DocLayoutV3PostProcess
  transformer: DocLayoutV3Transformer
DocLayoutV3Head:
  loss:
    aux_loss: true
    loss_coeff:
      bbox: 5
      class: 4
      dice: 5
      giou: 2
      mask: 5
      order: 50
    matcher:
      matcher_coeff:
        bbox: 5
        class: 4
        dice: 5
        giou: 2
        mask: 5
      name: HungarianMatcher
    name: DocLayoutV3Loss
    use_focal_loss: true
    use_vfl: true
    vfl_iou_type: mask
DocLayoutV3Metric:
  eval_mask: false
DocLayoutV3PostProcess:
  mask_stride: 4
  num_top_queries: 300
  resize_mask: false
DocLayoutV3Transformer:
  activation: relu
  box_noise_scale: 1.0
  dim_feedforward: 1024
  dropout: 0.0
  feat_strides:
  - 8
  - 16
  - 32
  label_noise_ratio: 0.5
  learnt_init_query: false
  mask_enhanced: true
  nhead: 8
  num_denoising: 0
  num_levels: 3
  num_queries: 300
  position_embed_type: sine
EvalDataset:
  anno_path: annotations/instance_val.json
  dataset_dir: /root/work/paddleocr-det/PaddleX/paddlex/dataset/handwriting
  image_dir: images
  name: COCOInstSegDataset
EvalReader:
  batch_size: 1
  drop_last: false
  sample_transforms:
  - Decode: {}
  - Resize:
      interp: 2
      keep_ratio: false
      target_size:
      - 800
      - 800
  - NormalizeImage:
      mean:
      - 0.0
      - 0.0
      - 0.0
      norm_type: none
      std:
      - 1.0
      - 1.0
      - 1.0
  - Permute: {}
  shuffle: false
LearningRate:
  base_lr: 0.0001
  schedulers:
  - !PiecewiseDecay
    gamma:
    - 1.0
    milestones:
    - 100
    use_warmup: true
    values: null
  - !LinearWarmup
    epochs: null
    epochs_first: false
    start_factor: 0.001
    steps: 200
MaskHybridEncoder:
  encoder_layer:
    activation: gelu
    d_model: 256
    dim_feedforward: 1024
    dropout: 0.0
    name: TransformerLayer
    nhead: 8
  expansion: 1.0
  hidden_dim: 256
  mask_feat_channels:
  - 64
  - 64
  num_encoder_layers: 1
  use_encoder_idx:
  - 3
OptimizerBuilder:
  clip_grad_by_norm: 0.1
  optimizer:
    type: AdamW
    weight_decay: 0.0001
  regularizer: false
PPHGNetV2:
  arch: L
  freeze_at: 0
  freeze_norm: true
  freeze_stem_only: true
  lr_mult_list:
  - 0.0
  - 0.05
  - 0.05
  - 0.05
  - 0.05
  return_idx:
  - 0
  - 1
  - 2
  - 3
TestDataset: !ImageFolder
  anno_path: annotations/instance_val.json
  dataset_dir: /root/work/paddleocr-det/PaddleX/paddlex/dataset/handwriting
  image_dir: ''
  sample_num: -1
  use_default_label: null
TestReader:
  batch_size: 1
  drop_last: false
  inputs_def:
    image_shape:
    - 3
    - 800
    - 800
  sample_transforms:
  - Decode: {}
  - Resize:
      interp: 2
      keep_ratio: false
      target_size:
      - 800
      - 800
  - NormalizeImage:
      mean:
      - 0.0
      - 0.0
      - 0.0
      norm_type: none
      std:
      - 1.0
      - 1.0
      - 1.0
  - Permute: {}
  shuffle: false
TrainDataset:
  anno_path: annotations/instance_train.json
  data_fields:
  - image
  - gt_bbox
  - gt_class
  - gt_poly
  - is_crowd
  - gt_read_order
  dataset_dir: /root/work/paddleocr-det/PaddleX/paddlex/dataset/handwriting
  image_dir: images
  name: COCOInstSegDataset
TrainReader:
  batch_size: 8
  batch_transforms:
  - BatchRandomResize:
      keep_ratio: false
      random_interp: true
      random_size: true
      target_size:
      - 672
      - 704
      - 736
      - 768
      - 800
      - 800
      - 800
      - 800
      - 832
      - 864
      - 896
      - 928
  - UnpackMask: {}
  - NormalizeImage:
      mean:
      - 0.0
      - 0.0
      - 0.0
      norm_type: none
      std:
      - 1.0
      - 1.0
      - 1.0
  - NormalizeBox: {}
  - BboxXYXY2XYWH: {}
  - Permute: {}
  collate_batch: false
  drop_last: true
  sample_transforms:
  - Decode: {}
  - Poly2MaskPack:
      del_poly: true
  - RandomDistort:
      prob: 0.8
  - UpdateBBoxFromMask: {}
  - RandomExpand:
      fill_value:
      - 123.675
      - 116.28
      - 103.53
      prob: 0.5
      ratio: 1.5
  - RandomCrop:
      prob: 0.8
      use_box_candidates: true
  shuffle: true
  use_shared_memory: true
amp: false
architecture: DETR
ema_decay: 0.9999
ema_decay_type: exponential
ema_filter_no_grad: true
enable_ce: false
epoch: 200
eval: true
eval_size:
- 800
- 800
export:
  benchmark: false
  fuse_conv_bn: false
  nms: true
  post_process: true
export_with_pir: true
filename: instancesegmodel_PP-DocLayoutV3
find_unused_parameters: true
fleet: false
hidden_dim: 256
log_iter: 10
metric: DocLayoutV3Metric
norm_type: sync_bn
num_classes: 6
num_prototypes: 32
pdx_model_name: PP-DocLayoutV3
pretrain_weights: /root/work/paddleocr-det/PaddleX/PP-DocLayoutV3_pretrained.pdparams
print_flops: false
print_params: false
profiler_options: null
proposals_path: sniper/proposals.json
resume: null
save_dir: /root/work/paddleocr-det/PaddleX/output
save_prediction_only: false
save_proposals: false
snapshot_epoch: 5
target_metrics: mask
to_static: false
uniform_output_enabled: true
use_ema: true
use_focal_loss: true
use_gpu: true
use_iluvatar_gpu: false
use_mlu: false
use_npu: false
use_vdl: true
use_wandb: false
use_xpu: false
vdl_log_dir: /root/work/paddleocr-det/PaddleX/output
with_mask: true
worker_num: 24