# Copyright (C) 2020-2022, François-Guillaume Fernandez.
# This program is licensed under the Apache License 2.0.
# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0> for full license details.
from collections import OrderedDict
from typing import Any, Callable, Dict, List, Optional
import torch.nn as nn
from holocron.nn import GlobalAvgPool2d
from holocron.nn.init import init_module
from ..presets import IMAGENETTE
from ..utils import conv_sequence, load_pretrained_params
__all__ = ["DarknetV1", "darknet24"]
default_cfgs: Dict[str, Dict[str, Any]] = {
"darknet24": {
**IMAGENETTE,
"input_shape": (3, 224, 224),
"url": "https://github.com/frgfm/Holocron/releases/download/v0.1.3/darknet24_224-816d72cb.pt",
},
}
class DarknetBodyV1(nn.Sequential):
def __init__(
self,
layout: List[List[int]],
in_channels: int = 3,
stem_channels: int = 64,
act_layer: Optional[nn.Module] = None,
norm_layer: Optional[Callable[[int], nn.Module]] = None,
drop_layer: Optional[Callable[..., nn.Module]] = None,
conv_layer: Optional[Callable[..., nn.Module]] = None,
) -> None:
if act_layer is None:
act_layer = nn.LeakyReLU(0.1, inplace=True)
in_chans = [stem_channels] + [_layout[-1] for _layout in layout[:-1]]
super().__init__(
OrderedDict(
[
(
"stem",
nn.Sequential(
*conv_sequence(
in_channels,
stem_channels,
act_layer,
norm_layer,
drop_layer,
conv_layer,
kernel_size=7,
padding=3,
stride=2,
bias=(norm_layer is None),
)
),
),
(
"layers",
nn.Sequential(
*[
self._make_layer([_in_chans] + planes, act_layer, norm_layer, drop_layer, conv_layer)
for _in_chans, planes in zip(in_chans, layout)
]
),
),
]
)
)
@staticmethod
def _make_layer(
planes: List[int],
act_layer: Optional[nn.Module] = None,
norm_layer: Optional[Callable[[int], nn.Module]] = None,
drop_layer: Optional[Callable[..., nn.Module]] = None,
conv_layer: Optional[Callable[..., nn.Module]] = None,
) -> nn.Sequential:
_layers: List[nn.Module] = [nn.MaxPool2d(2)]
k1 = True
for in_planes, out_planes in zip(planes[:-1], planes[1:]):
_layers.extend(
conv_sequence(
in_planes,
out_planes,
act_layer,
norm_layer,
drop_layer,
conv_layer,
kernel_size=3 if out_planes > in_planes else 1,
padding=1 if out_planes > in_planes else 0,
bias=(norm_layer is None),
)
)
k1 = not k1
return nn.Sequential(*_layers)
class DarknetV1(nn.Sequential):
def __init__(
self,
layout: List[List[int]],
num_classes: int = 10,
in_channels: int = 3,
stem_channels: int = 64,
act_layer: Optional[nn.Module] = None,
norm_layer: Optional[Callable[[int], nn.Module]] = None,
drop_layer: Optional[Callable[..., nn.Module]] = None,
conv_layer: Optional[Callable[..., nn.Module]] = None,
) -> None:
super().__init__(
OrderedDict(
[
(
"features",
DarknetBodyV1(
layout, in_channels, stem_channels, act_layer, norm_layer, drop_layer, conv_layer
),
),
("pool", GlobalAvgPool2d(flatten=True)),
("classifier", nn.Linear(layout[2][-1], num_classes)),
]
)
)
init_module(self, "leaky_relu")
def _darknet(arch: str, pretrained: bool, progress: bool, layout: List[List[int]], **kwargs: Any) -> DarknetV1:
# Build the model
model = DarknetV1(layout, **kwargs)
model.default_cfg = default_cfgs[arch] # type: ignore[assignment]
# Load pretrained parameters
if pretrained:
load_pretrained_params(model, default_cfgs[arch]["url"], progress)
return model
[docs]
def darknet24(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> DarknetV1:
"""Darknet-24 from
`"You Only Look Once: Unified, Real-Time Object Detection" <https://pjreddie.com/media/files/papers/yolo_1.pdf>`_
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
progress (bool): If True, displays a progress bar of the download to stderr
Returns:
torch.nn.Module: classification model
"""
return _darknet(
"darknet24",
pretrained,
progress,
[[192], [128, 256, 256, 512], [*([256, 512] * 4), 512, 1024], [512, 1024] * 2],
**kwargs,
)