157 lines
8 KiB
Python
157 lines
8 KiB
Python
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
|
|
# This file was automatically generated from src/transformers/models/lightglue/modular_lightglue.py.
|
|
# Do NOT edit this file manually as any edits will be overwritten by the generation of
|
|
# the file from the modular. If any change should be done, please apply the change to the
|
|
# modular_lightglue.py file directly. One of our CI enforces this.
|
|
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
|
|
# Copyright 2025 The HuggingFace Team. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
|
|
from ...configuration_utils import PretrainedConfig
|
|
from ..auto import CONFIG_MAPPING, AutoConfig
|
|
from ..superpoint import SuperPointConfig
|
|
|
|
|
|
class LightGlueConfig(PretrainedConfig):
|
|
r"""
|
|
This is the configuration class to store the configuration of a [`LightGlueForKeypointMatching`]. It is used to
|
|
instantiate a LightGlue 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 LightGlue
|
|
[ETH-CVG/lightglue_superpoint](https://huggingface.co/ETH-CVG/lightglue_superpoint) architecture.
|
|
|
|
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
|
|
documentation from [`PretrainedConfig`] for more information.
|
|
|
|
Args:
|
|
keypoint_detector_config (`Union[AutoConfig, dict]`, *optional*, defaults to `SuperPointConfig`):
|
|
The config object or dictionary of the keypoint detector.
|
|
descriptor_dim (`int`, *optional*, defaults to 256):
|
|
The dimension of the descriptors.
|
|
num_hidden_layers (`int`, *optional*, defaults to 9):
|
|
The number of self and cross attention layers.
|
|
num_attention_heads (`int`, *optional*, defaults to 4):
|
|
The number of heads in the multi-head attention.
|
|
num_key_value_heads (`int`, *optional*):
|
|
This is the number of key_value heads that should be used to implement Grouped Query Attention. If
|
|
`num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
|
|
`num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
|
|
converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
|
|
by meanpooling all the original heads within that group. For more details checkout [this
|
|
paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
|
|
`num_attention_heads`.
|
|
depth_confidence (`float`, *optional*, defaults to 0.95):
|
|
The confidence threshold used to perform early stopping
|
|
width_confidence (`float`, *optional*, defaults to 0.99):
|
|
The confidence threshold used to prune points
|
|
filter_threshold (`float`, *optional*, defaults to 0.1):
|
|
The confidence threshold used to filter matches
|
|
initializer_range (`float`, *optional*, defaults to 0.02):
|
|
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
|
|
hidden_act (`str`, *optional*, defaults to `"gelu"`):
|
|
The activation function to be used in the hidden layers.
|
|
attention_dropout (`float`, *optional*, defaults to 0.0):
|
|
The dropout ratio for the attention probabilities.
|
|
attention_bias (`bool`, *optional*, defaults to `True`):
|
|
Whether to use a bias in the query, key, value and output projection layers during self-attention.
|
|
trust_remote_code (`bool`, *optional*, defaults to `False`):
|
|
Whether to trust remote code when using other models than SuperPoint as keypoint detector.
|
|
|
|
Examples:
|
|
```python
|
|
>>> from transformers import LightGlueConfig, LightGlueForKeypointMatching
|
|
|
|
>>> # Initializing a LightGlue style configuration
|
|
>>> configuration = LightGlueConfig()
|
|
|
|
>>> # Initializing a model from the LightGlue style configuration
|
|
>>> model = LightGlueForKeypointMatching(configuration)
|
|
|
|
>>> # Accessing the model configuration
|
|
>>> configuration = model.config
|
|
```
|
|
"""
|
|
|
|
model_type = "lightglue"
|
|
sub_configs = {"keypoint_detector_config": AutoConfig}
|
|
|
|
def __init__(
|
|
self,
|
|
keypoint_detector_config: SuperPointConfig = None,
|
|
descriptor_dim: int = 256,
|
|
num_hidden_layers: int = 9,
|
|
num_attention_heads: int = 4,
|
|
num_key_value_heads=None,
|
|
depth_confidence: float = 0.95,
|
|
width_confidence: float = 0.99,
|
|
filter_threshold: float = 0.1,
|
|
initializer_range: float = 0.02,
|
|
hidden_act: str = "gelu",
|
|
attention_dropout=0.0,
|
|
attention_bias=True,
|
|
trust_remote_code: bool = False,
|
|
**kwargs,
|
|
):
|
|
# LightGlue can be used with other models than SuperPoint as keypoint detector
|
|
# We provide the trust_remote_code argument to allow the use of other models
|
|
# that are not registered in the CONFIG_MAPPING dictionary (for example DISK)
|
|
self.trust_remote_code = trust_remote_code
|
|
|
|
if descriptor_dim % num_attention_heads != 0:
|
|
raise ValueError("descriptor_dim % num_heads is different from zero")
|
|
|
|
self.descriptor_dim = descriptor_dim
|
|
self.num_hidden_layers = num_hidden_layers
|
|
self.num_attention_heads = num_attention_heads
|
|
|
|
# for backward compatibility
|
|
if num_key_value_heads is None:
|
|
num_key_value_heads = num_attention_heads
|
|
|
|
self.num_key_value_heads = num_key_value_heads
|
|
|
|
self.depth_confidence = depth_confidence
|
|
self.width_confidence = width_confidence
|
|
self.filter_threshold = filter_threshold
|
|
self.initializer_range = initializer_range
|
|
|
|
# Keypoint Detector is forced into eager attention mode because SuperPoint does not have Attention
|
|
# See https://github.com/huggingface/transformers/pull/31718#discussion_r2109733153
|
|
if isinstance(keypoint_detector_config, dict):
|
|
keypoint_detector_config["model_type"] = (
|
|
keypoint_detector_config["model_type"] if "model_type" in keypoint_detector_config else "superpoint"
|
|
)
|
|
if keypoint_detector_config["model_type"] not in CONFIG_MAPPING:
|
|
keypoint_detector_config = AutoConfig.from_pretrained(
|
|
keypoint_detector_config["_name_or_path"], trust_remote_code=self.trust_remote_code
|
|
)
|
|
else:
|
|
keypoint_detector_config = CONFIG_MAPPING[keypoint_detector_config["model_type"]](
|
|
**keypoint_detector_config, attn_implementation="eager"
|
|
)
|
|
|
|
if keypoint_detector_config is None:
|
|
keypoint_detector_config = CONFIG_MAPPING["superpoint"](attn_implementation="eager")
|
|
|
|
self.keypoint_detector_config = keypoint_detector_config
|
|
|
|
self.hidden_size = descriptor_dim
|
|
self.intermediate_size = descriptor_dim * 2
|
|
self.hidden_act = hidden_act
|
|
self.attention_dropout = attention_dropout
|
|
self.attention_bias = attention_bias
|
|
super().__init__(**kwargs)
|
|
|
|
|
|
__all__ = ["LightGlueConfig"]
|