File size: 4,427 Bytes
49f3b67 00797c3 49f3b67 00797c3 49f3b67 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 |
# Copyright (c) 2023, NVIDIA CORPORATION. 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 collections import namedtuple
from typing import Optional
from einops import rearrange
from timm.models import VisionTransformer
import torch
from transformers import PretrainedConfig, PreTrainedModel
from .eradio_model import eradio
from .radio_model import create_model_from_args
from .radio_model import RADIOModel as RADIOModelBase
from .input_conditioner import get_default_conditioner, InputConditioner
class RADIOConfig(PretrainedConfig):
"""Pretrained Hugging Face configuration for RADIO models."""
def __init__(
self,
args: Optional[dict] = None,
version: Optional[str] = "v1",
return_summary: Optional[bool] = True,
return_spatial_features: Optional[bool] = True,
**kwargs,
):
self.args = args
self.version = version
self.return_summary = return_summary
self.return_spatial_features = return_spatial_features
super().__init__(**kwargs)
class RADIOModel(PreTrainedModel):
"""Pretrained Hugging Face model for RADIO.
This class inherits from PreTrainedModel, which provides
HuggingFace's functionality for loading and saving models.
"""
config_class = RADIOConfig
def __init__(self, config):
super().__init__(config)
RADIOArgs = namedtuple("RADIOArgs", config.args.keys())
args = RADIOArgs(**config.args)
self.config = config
model = create_model_from_args(args)
input_conditioner: InputConditioner = get_default_conditioner()
self.radio_model = RADIOModelBase(
model,
input_conditioner,
config.return_summary,
config.return_spatial_features,
)
@property
def model(self) -> VisionTransformer:
return self.radio_model.model
@property
def input_conditioner(self) -> InputConditioner:
return self.radio_model.input_conditioner
def forward(self, x: torch.Tensor):
return self.radio_model.forward(x)
class ERADIOConfig(PretrainedConfig):
"""Pretrained Hugging Face configuration for ERADIO models."""
def __init__(
self,
args: Optional[dict] = None,
version: Optional[str] = "v1",
return_summary: Optional[bool] = True,
return_spatial_features: Optional[bool] = True,
**kwargs,
):
self.args = args
self.version = version
self.return_summary = return_summary
self.return_spatial_features = return_spatial_features
super().__init__(**kwargs)
class ERADIOModel(PreTrainedModel):
"""Pretrained Hugging Face model for ERADIO.
This class inherits from PreTrainedModel, which provides
HuggingFace's functionality for loading and saving models.
"""
config_class = ERADIOConfig
def __init__(self, config):
super().__init__(config)
config.args["in_chans"] = 3
config.args["num_classes"] = 0
config.args["return_full_features"] = config.return_spatial_features
self.config = config
model = eradio(**config.args)
self.input_conditioner: InputConditioner = get_default_conditioner()
self.return_summary = config.return_summary
self.return_spatial_features = config.return_spatial_features
self.model = model
def forward(self, x: torch.Tensor):
x = self.input_conditioner(x)
y = self.model.forward_features(x)
summary, features = self.model.forward_features(x)
if isinstance(y, tuple):
summary, features = y
else:
summary = y
features = None
if self.return_summary and self.return_spatial_features:
return summary, features
elif self.return_summary:
return summary
return features
|