a h(@s>ddlmZddlmZddlmZGdddeZdgZdS))Optional)PretrainedConfig)rope_config_validationcszeZdZdZdZdeeeeeeeeeeeeeeeeee e ee e e eee e e ee ee e dfdd Z Z S)EfficientLoFTRConfiga  This is the configuration class to store the configuration of a [`EffientLoFTRFromKeypointMatching`]. It is used to instantiate a EfficientLoFTR model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the EfficientLoFTR [zju-community/efficientloftr](https://huggingface.co/zju-community/efficientloftr) architecture. Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: stage_num_blocks (`List`, *optional*, defaults to [1, 2, 4, 14]): The number of blocks in each stages out_features (`List`, *optional*, defaults to [64, 64, 128, 256]): The number of channels in each stage stage_stride (`List`, *optional*, defaults to [2, 1, 2, 2]): The stride used in each stage hidden_size (`int`, *optional*, defaults to 256): The dimension of the descriptors. activation_function (`str`, *optional*, defaults to `"relu"`): The activation function used in the backbone q_aggregation_kernel_size (`int`, *optional*, defaults to 4): The kernel size of the aggregation of query states in the fusion network kv_aggregation_kernel_size (`int`, *optional*, defaults to 4): The kernel size of the aggregation of key and value states in the fusion network q_aggregation_stride (`int`, *optional*, defaults to 4): The stride of the aggregation of query states in the fusion network kv_aggregation_stride (`int`, *optional*, defaults to 4): The stride of the aggregation of key and value states in the fusion network num_attention_layers (`int`, *optional*, defaults to 4): Number of attention layers in the LocalFeatureTransformer num_attention_heads (`int`, *optional*, defaults to 8): The number of heads in the GNN layers. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. attention_bias (`bool`, *optional*, defaults to `False`): Whether to use a bias in the query, key, value and output projection layers during attention. mlp_activation_function (`str`, *optional*, defaults to `"leaky_relu"`): Activation function used in the attention mlp layer. coarse_matching_skip_softmax (`bool`, *optional*, defaults to `False`): Whether to skip softmax or not at the coarse matching step. coarse_matching_threshold (`float`, *optional*, defaults to 0.2): The threshold for the minimum score required for a match. coarse_matching_temperature (`float`, *optional*, defaults to 0.1): The temperature to apply to the coarse similarity matrix coarse_matching_border_removal (`int`, *optional*, defaults to 2): The size of the border to remove during coarse matching fine_kernel_size (`int`, *optional*, defaults to 8): Kernel size used for the fine feature matching batch_norm_eps (`float`, *optional*, defaults to 1e-05): The epsilon used by the batch normalization layers. rope_theta (`float`, *optional*, defaults to 10000.0): The base period of the RoPE embeddings. partial_rotary_factor (`float`, *optional*, defaults to 4.0): Dim factor for the RoPE embeddings, in EfficientLoFTR, frequencies should be generated for the whole hidden_size, so this factor is used to compensate. rope_scaling (`Dict`, *optional*): Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value accordingly. Expected contents: `rope_type` (`str`): The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope', 'llama3', '2d'], with 'default' being the original RoPE implementation. `dim` (`int`): The dimension of the RoPE embeddings. fine_matching_slice_dim (`int`, *optional*, defaults to 8): The size of the slice used to divide the fine features for the first and second fine matching stages. fine_matching_regress_temperature (`float`, *optional*, defaults to 10.0): The temperature to apply to the fine similarity matrix initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. Examples: ```python >>> from transformers import EfficientLoFTRConfig, EfficientLoFTRForKeypointMatching >>> # Initializing a EfficientLoFTR configuration >>> configuration = EfficientLoFTRConfig() >>> # Initializing a model from the EfficientLoFTR configuration >>> model = EfficientLoFTRForKeypointMatching(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` ZefficientloftrNreluF leaky_relu皙?皙?h㈵>@@$@{Gz?)stage_num_blocks out_features stage_stride hidden_sizeactivation_functionq_aggregation_kernel_sizekv_aggregation_kernel_sizeq_aggregation_stridekv_aggregation_stridenum_attention_layersnum_attention_headsattention_dropoutattention_biasmlp_activation_functioncoarse_matching_skip_softmaxcoarse_matching_thresholdcoarse_matching_temperaturecoarse_matching_border_removalfine_kernel_sizebatch_norm_eps rope_thetapartial_rotary_factor rope_scalingfine_matching_slice_dim!fine_matching_regress_temperatureinitializer_rangec s|dur |ngd_|dur"|ngd_|dur8|ngd_dgjdd_ddtjjD_fddtjD_fd dtt jD_ t t jdd_ |_jjdkrtd jd jd|_|_|_|_| _| _| _| _| _jd _|_|_|_|_|_|_|_ |_!|_"| _#|_$|dur|nd di_%|_&t'|_(t)j*fi|dS)N)rr )rr/rr)@r1rr/cSs$g|]\}}|gdg|dqS)r/).0Zstride num_blocksr4r4{/var/www/html/assistant/venv/lib/python3.9/site-packages/transformers/models/efficientloftr/configuration_efficientloftr.py sz1EfficientLoFTRConfig.__init__..cs g|]\}}j|g|qSr4)r)r5 stage_idxr6selfr4r7r8scs*g|]"}j|gj|ddqS)Nr3)stage_in_channelsstage_block_out_channels)r5r9r:r4r7r8szMhidden_size should be equal to the last value in out_features. hidden_size = z, out_features = rZ rope_typedefault)+rrrr<zipZstage_block_stride enumerater=rangelenZstage_block_in_channelslistreversedZfine_fusion_dimsr ValueErrorrrrrrrrr r!Zintermediate_sizer"r#r$r%r&r'r(r,r-Znum_key_value_headsr)r+r*rr.super__init__)r;rrrrrrrrrrrr r!r"r#r$r%r&r'r(r)r*r+r,r-r.kwargs __class__r:r7rGmsZ     zEfficientLoFTRConfig.__init__)NNNrrr r r r r r r Fr Fr rrr rrrNr rr)__name__ __module__ __qualname____doc__Z model_typerrCintstrfloatbooldictrG __classcell__r4r4rIr7rspV   rN)typingrZconfiguration_utilsrZmodeling_rope_utilsrr__all__r4r4r4r7s   4