config.json

{
  "activation_dropout": 0.0,
  "activation_function": "silu",
  "anchor_image_size": null,
  "architectures": [
    "DFineForObjectDetection"
  ],
  "attention_dropout": 0.0,
  "auxiliary_loss": true,
  "backbone": null,
  "backbone_config": {
    "depths": [
      3,
      4,
      6,
      3
    ],
    "downsample_in_bottleneck": false,
    "downsample_in_first_stage": false,
    "dtype": "float32",
    "embedding_size": 32,
    "hidden_act": "relu",
    "hidden_sizes": [
      128,
      256,
      512,
      1024
    ],
    "initializer_range": 0.02,
    "layer_type": "basic",
    "model_type": "hgnet_v2",
    "num_channels": 3,
    "out_features": [
      "stage2",
      "stage3",
      "stage4"
    ],
    "out_indices": [
      2,
      3,
      4
    ],
    "stage_downsample": [
      false,
      true,
      true,
      true
    ],
    "stage_in_channels": [
      16,
      64,
      256,
      512
    ],
    "stage_kernel_size": [
      3,
      3,
      5,
      5
    ],
    "stage_light_block": [
      false,
      false,
      true,
      true
    ],
    "stage_mid_channels": [
      16,
      32,
      64,
      128
    ],
    "stage_names": [
      "stem",
      "stage1",
      "stage2",
      "stage3",
      "stage4"
    ],
    "stage_num_blocks": [
      1,
      1,
      2,
      1
    ],
    "stage_numb_of_layers": [
      3,
      3,
      3,
      3
    ],
    "stage_out_channels": [
      64,
      256,
      512,
      1024
    ],
    "stem_channels": [
      3,
      16,
      16
    ],
    "use_learnable_affine_block": true
  },
  "batch_norm_eps": 1e-05,
  "box_noise_scale": 1.0,
  "d_model": 256,
  "decoder_activation_function": "relu",
  "decoder_attention_heads": 8,
  "decoder_ffn_dim": 1024,
  "decoder_in_channels": [
    256,
    256,
    256
  ],
  "decoder_layers": 3,
  "decoder_method": "default",
  "decoder_n_points": [
    3,
    6,
    3
  ],
  "decoder_offset_scale": 0.5,
  "depth_mult": 0.34,
  "dropout": 0.0,
  "dtype": "float32",
  "encode_proj_layers": [
    2
  ],
  "encoder_activation_function": "gelu",
  "encoder_attention_heads": 8,
  "encoder_ffn_dim": 1024,
  "encoder_hidden_dim": 256,
  "encoder_in_channels": [
    256,
    512,
    1024
  ],
  "encoder_layers": 1,
  "eos_coefficient": 0.0001,
  "eval_idx": -1,
  "eval_size": null,
  "feat_strides": [
    8,
    16,
    32
  ],
  "focal_loss_alpha": 0.75,
  "focal_loss_gamma": 2.0,
  "freeze_backbone_batch_norms": true,
  "hidden_expansion": 0.5,
  "id2label": {
    "0": "0",
    "1": "1",
    "2": "2",
    "3": "3",
    "4": "4"
  },
  "initializer_bias_prior_prob": null,
  "initializer_range": 0.01,
  "is_encoder_decoder": true,
  "label2id": {
    "0": 0,
    "1": 1,
    "2": 2,
    "3": 3,
    "4": 4
  },
  "label_noise_ratio": 0.5,
  "layer_norm_eps": 1e-05,
  "layer_scale": 1,
  "learn_initial_query": false,
  "lqe_hidden_dim": 64,
  "lqe_layers": 2,
  "matcher_alpha": 0.25,
  "matcher_bbox_cost": 5.0,
  "matcher_class_cost": 2.0,
  "matcher_gamma": 2.0,
  "matcher_giou_cost": 2.0,
  "max_num_bins": 32,
  "model_type": "d_fine",
  "normalize_before": false,
  "num_denoising": 100,
  "num_feature_levels": 3,
  "num_queries": 300,
  "positional_encoding_temperature": 10000,
  "reg_scale": 4.0,
  "tie_word_embeddings": true,
  "top_prob_values": 4,
  "transformers_version": "5.3.0",
  "up": 0.5,
  "use_focal_loss": true,
  "weight_loss_bbox": 5.0,
  "weight_loss_ddf": 1.5,
  "weight_loss_fgl": 0.15,
  "weight_loss_giou": 2.0,
  "weight_loss_vfl": 1.0,
  "with_box_refine": true
}