PaddleX/paddlex/repo_apis/PaddleDetection_api/configs/Mask-RT-DETR-L.yaml

# Runtime
epoch: 72
use_gpu: true
use_xpu: false
use_mlu: false
use_npu: false
log_iter: 20
save_dir: output
snapshot_epoch: 1
target_metrics: mask
print_flops: false
print_params: false
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
ema_filter_no_grad: True

# Dataset
metric: COCO
num_classes: 80

worker_num: 4

TrainDataset:
  name: COCOInstSegDataset
  image_dir: images
  anno_path: annotations/instance_train.json
  dataset_dir: datasets/COCO
  data_fields: ['image', 'gt_bbox', 'gt_class', 'gt_poly', 'is_crowd']

EvalDataset:
  name: COCOInstSegDataset
  image_dir: images
  anno_path: annotations/instance_val.json
  dataset_dir: datasets/COCO

TestDataset:
  name: ImageFolder
  anno_path: annotations/instance_val.json
  dataset_dir: datasets/COCO

TrainReader:
  sample_transforms:
    - Decode: {}
    - Poly2Mask: {del_poly: True}
    - RandomDistort: {prob: 0.8}
    - RandomExpand: {fill_value: [123.675, 116.28, 103.53]}
    - RandomCrop: {prob: 0.8}
    - RandomFlip: {}
  batch_transforms:
    - BatchRandomResize: {target_size: [480, 512, 544, 576, 608, 640, 640, 640, 672, 704, 736, 768, 800], random_size: True, random_interp: True, keep_ratio: False}
    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
    - NormalizeBox: {}
    - BboxXYXY2XYWH: {}
    - Permute: {}
  batch_size: 4
  shuffle: true
  drop_last: true
  collate_batch: false
  use_shared_memory: true

EvalReader:
  sample_transforms:
    - Decode: {}
    - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}
    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
    - Permute: {}
  batch_size: 1 # mask be 1
  shuffle: false
  drop_last: false

TestReader:
  inputs_def:
    image_shape: [3, 640, 640]
  sample_transforms:
    - Decode: {}
    - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}
    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
    - Permute: {}
  batch_size: 1
  shuffle: false
  drop_last: false

LearningRate:
  base_lr: 0.0001
  schedulers:
  - !PiecewiseDecay
    gamma: 1.0
    milestones: [100]
    use_warmup: true
  - !LinearWarmup
    start_factor: 0.001
    steps: 2000

OptimizerBuilder:
  clip_grad_by_norm: 0.1
  regularizer: false
  optimizer:
    type: AdamW
    weight_decay: 0.0001

# Model
architecture: DETR
with_mask: True
pretrain_weights: https://bj.bcebos.com/v1/paddledet/models/pretrained/PPHGNetV2_L_ssld_pretrained.pdparams
norm_type: sync_bn
hidden_dim: 256
use_focal_loss: True
eval_size: [640, 640]
num_prototypes: 32
find_unused_parameters: True

DETR:
  backbone: PPHGNetV2
  neck: MaskHybridEncoder
  transformer: MaskRTDETR
  detr_head: MaskDINOHead
  post_process: DETRPostProcess

PPHGNetV2:
  arch: 'L'
  return_idx: [0, 1, 2, 3]
  freeze_stem_only: True
  freeze_at: 0
  freeze_norm: True
  lr_mult_list: [0., 0.05, 0.05, 0.05, 0.05]

MaskHybridEncoder:
  hidden_dim: 256
  use_encoder_idx: [3]
  num_encoder_layers: 1
  encoder_layer:
    name: TransformerLayer
    d_model: 256
    nhead: 8
    dim_feedforward: 1024
    dropout: 0.
    activation: 'gelu'
  expansion: 1.0
  mask_feat_channels: [64, 64]

MaskRTDETR:
  num_queries: 300
  position_embed_type: sine
  feat_strides: [8, 16, 32]
  num_levels: 3
  nhead: 8
  dim_feedforward: 1024
  dropout: 0.0
  activation: relu
  num_denoising: 100
  label_noise_ratio: 0.5
  box_noise_scale: 1.0
  learnt_init_query: False
  mask_enhanced: True

MaskDINOHead:
  loss:
    name: MaskDINOLoss
    loss_coeff: {class: 4, bbox: 5, giou: 2, mask: 5, dice: 5}
    aux_loss: True
    use_vfl: True
    vfl_iou_type: 'mask'
    matcher:
      name: HungarianMatcher
      matcher_coeff: {class: 4, bbox: 5, giou: 2, mask: 5, dice: 5}

DETRPostProcess:
  num_top_queries: 100
  mask_stride: 4

# Exporting the model
export:
  post_process: True
  nms: True
  benchmark: False
  fuse_conv_bn: False