---
library_name: keras-hub
---
This is a [`DFine` model](https://keras.io/api/keras_hub/models/d_fine) uploaded using the KerasHub library and can be used with JAX, TensorFlow, and PyTorch backends.
This model is related to a `ObjectDetector` task.

Model config:
* **name:** d_fine_backbone
* **trainable:** True
* **backbone:** {'module': 'keras_hub.src.models.hgnetv2.hgnetv2_backbone', 'class_name': 'HGNetV2Backbone', 'config': {'name': 'hg_net_v2_backbone', 'trainable': True, 'depths': [3, 4, 6, 3], 'embedding_size': 32, 'hidden_sizes': [256, 512, 1024, 2048], 'stem_channels': [3, 32, 64], 'hidden_act': 'relu', 'use_learnable_affine_block': False, 'stackwise_stage_filters': [[64, 64, 128, 1, 6, 3], [128, 128, 512, 2, 6, 3], [512, 256, 1024, 5, 6, 5], [1024, 512, 2048, 2, 6, 5]], 'apply_downsample': [False, True, True, True], 'use_lightweight_conv_block': [False, False, True, True], 'image_shape': [None, None, 3], 'out_features': ['stage2', 'stage3', 'stage4'], 'data_format': 'channels_last'}, 'registered_name': 'keras_hub>HGNetV2Backbone'}
* **decoder_in_channels:** [384, 384, 384]
* **encoder_hidden_dim:** 384
* **num_labels:** 366
* **num_denoising:** 100
* **learn_initial_query:** False
* **num_queries:** 300
* **anchor_image_size:** [640, 640]
* **feat_strides:** [8, 16, 32]
* **num_feature_levels:** 3
* **hidden_dim:** 256
* **encoder_in_channels:** [512, 1024, 2048]
* **encode_proj_layers:** [2]
* **num_attention_heads:** 8
* **encoder_ffn_dim:** 2048
* **num_encoder_layers:** 1
* **hidden_expansion:** 1.0
* **depth_multiplier:** 1.0
* **eval_idx:** -1
* **box_noise_scale:** 1.0
* **label_noise_ratio:** 0.5
* **labels:** None
* **num_decoder_layers:** 6
* **decoder_attention_heads:** 8
* **decoder_ffn_dim:** 1024
* **decoder_method:** default
* **decoder_n_points:** [3, 6, 3]
* **lqe_hidden_dim:** 64
* **num_lqe_layers:** 2
* **seed:** 0
* **image_shape:** [None, None, 3]
* **data_format:** channels_last
* **out_features:** ['stage2', 'stage3', 'stage4']

This model card has been generated automatically and should be completed by the model author. See [Model Cards documentation](https://huggingface.co/docs/hub/model-cards) for more information.