✨ [Add] v9-c-segment model, inference WIP
Browse files- yolo/model/module.py +32 -0
- yolo/model/yolo.py +1 -1
yolo/model/module.py
CHANGED
@@ -130,6 +130,38 @@ class MultiheadDetection(nn.Module):
|
|
130 |
return [head(x) for x, head in zip(x_list, self.heads)]
|
131 |
|
132 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
133 |
class Anchor2Vec(nn.Module):
|
134 |
def __init__(self, reg_max: int = 16) -> None:
|
135 |
super().__init__()
|
|
|
130 |
return [head(x) for x, head in zip(x_list, self.heads)]
|
131 |
|
132 |
|
133 |
+
class Segmentation(nn.Module):
|
134 |
+
def __init__(self, in_channels: Tuple[int], num_maskes: int):
|
135 |
+
super().__init__()
|
136 |
+
first_neck, in_channels = in_channels
|
137 |
+
|
138 |
+
mask_neck = max(first_neck // 4, num_maskes)
|
139 |
+
self.mask_conv = nn.Sequential(
|
140 |
+
Conv(in_channels, mask_neck, 3), Conv(mask_neck, mask_neck, 3), nn.Conv2d(mask_neck, num_maskes, 1)
|
141 |
+
)
|
142 |
+
|
143 |
+
def forward(self, x: Tensor) -> Tuple[Tensor]:
|
144 |
+
x = self.mask_conv(x)
|
145 |
+
return x
|
146 |
+
|
147 |
+
|
148 |
+
class MultiheadSegmentation(nn.Module):
|
149 |
+
"""Mutlihead Segmentation module for Dual segment or Triple segment"""
|
150 |
+
|
151 |
+
def __init__(self, in_channels: List[int], num_classes: int, num_maskes: int, **head_kwargs):
|
152 |
+
super().__init__()
|
153 |
+
mask_channels, proto_channels = in_channels[:-1], in_channels[-1]
|
154 |
+
|
155 |
+
self.detect = MultiheadDetection(mask_channels, num_classes, **head_kwargs)
|
156 |
+
self.heads = nn.ModuleList(
|
157 |
+
[Segmentation((in_channels[0], in_channel), num_maskes) for in_channel in mask_channels]
|
158 |
+
)
|
159 |
+
self.heads.append(Conv(proto_channels, num_maskes, 1))
|
160 |
+
|
161 |
+
def forward(self, x_list: List[torch.Tensor]) -> List[torch.Tensor]:
|
162 |
+
return [head(x) for x, head in zip(x_list, self.heads)]
|
163 |
+
|
164 |
+
|
165 |
class Anchor2Vec(nn.Module):
|
166 |
def __init__(self, reg_max: int = 16) -> None:
|
167 |
super().__init__()
|
yolo/model/yolo.py
CHANGED
@@ -45,7 +45,7 @@ class YOLO(nn.Module):
|
|
45 |
# Find in channels
|
46 |
if any(module in layer_type for module in ["Conv", "ELAN", "ADown", "AConv", "CBLinear"]):
|
47 |
layer_args["in_channels"] = output_dim[source]
|
48 |
-
if "Detection" in layer_type:
|
49 |
layer_args["in_channels"] = [output_dim[idx] for idx in source]
|
50 |
layer_args["num_classes"] = self.num_classes
|
51 |
layer_args["reg_max"] = self.reg_max
|
|
|
45 |
# Find in channels
|
46 |
if any(module in layer_type for module in ["Conv", "ELAN", "ADown", "AConv", "CBLinear"]):
|
47 |
layer_args["in_channels"] = output_dim[source]
|
48 |
+
if "Detection" in layer_type or "Segmentation" in layer_type:
|
49 |
layer_args["in_channels"] = [output_dim[idx] for idx in source]
|
50 |
layer_args["num_classes"] = self.num_classes
|
51 |
layer_args["reg_max"] = self.reg_max
|