[Application] darknet53 pytorch implementation for yolo v3

Added pytorch darknet53 model for yolo v3.

It is used in yolo v3 as a backbone model.

I'll add nntrainer implementation, too.

Signed-off-by: Seungbaek Hong <sb92.hong@samsung.com>

diff --git a/Applications/YOLOv3/PyTorch/yolo.py b/Applications/YOLOv3/PyTorch/yolo.py
new file mode 100644 (file)
index 0000000..7bb9f18
--- /dev/null
+++ b/Applications/YOLOv3/PyTorch/yolo.py
@@ -0,0 +1,90 @@
+#!/usr/bin/env python3
+# SPDX-License-Identifier: Apache-2.0
+# Copyright (C) 2023 Seungbaek Hong <sb92.hong@samsung.com>
+#
+# @file yolo.py
+# @date 31 May 2023
+# @brief Yolo v3 model
+#
+# @author Seungbaek Hong <sb92.hong@samsung.com>
+
+import torch
+import torch.nn as nn
+
+import sys
+import os
+
+# get path of pyutils using relative path
+def get_util_path():
+    current_path = os.path.abspath(os.path.dirname(__file__))
+    parent_path = os.path.abspath(os.path.dirname(current_path))
+    target_path = os.path.abspath(os.path.dirname(parent_path))
+    return os.path.dirname(target_path) + '/tools/pyutils/'
+
+# add path of pyutils to sys.path
+sys.path.append(get_util_path())
+from torchconverter import save_bin
+
+class ConvBlock(nn.Module):
+    def __init__(self, num_channel_in, num_channel_out, kernel_size, stride, padding):
+        super(ConvBlock, self).__init__()
+
+        self.conv = nn.Conv2d(num_channel_in, num_channel_out, kernel_size, stride, padding)
+        self.bn = nn.BatchNorm2d(num_channel_out, eps=1e-3)
+        self.leaky_relu = nn.LeakyReLU()
+    
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        x = self.leaky_relu(x)
+
+        return x
+
+
+class DarkNetBlock(nn.Module):
+    def __init__(self, num_channel, repeat=1):
+        super(DarkNetBlock, self).__init__()
+        self.repeat = repeat
+
+        self.module_list = nn.Sequential()
+        for _ in range(self.repeat):
+            self.module_list.append(nn.Sequential(
+                ConvBlock(num_channel, num_channel // 2, 1, 1, 0),
+                ConvBlock(num_channel // 2, num_channel, 3, 1, 1)
+            ))
+
+    def forward(self, x):        
+        for module in self.module_list:
+            out = module[0](x)
+            out = module[1](out)
+            x = x + out
+
+        return x
+
+class Darknet53(nn.Module):
+    def __init__(self):
+        super(Darknet53, self).__init__()
+
+        self.module_list = nn.Sequential()
+        self.module_list.append(ConvBlock(3, 32, 3, 1, 1))
+        self.module_list.append(ConvBlock(32, 64, 3, 2, 1))
+        self.module_list.append(DarkNetBlock(64, 1))
+        self.module_list.append(ConvBlock(64, 128, 3, 2, 1))
+        self.module_list.append(DarkNetBlock(128, 2))
+        self.module_list.append(ConvBlock(128, 256, 3, 2, 1))
+        self.module_list.append(DarkNetBlock(256, 8))
+        self.module_list.append(ConvBlock(256, 512, 3, 2, 1))
+        self.module_list.append(DarkNetBlock(512, 8))
+        self.module_list.append(ConvBlock(512, 1024, 3, 2, 1))
+        self.module_list.append(DarkNetBlock(1024, 4))        
+
+    def forward(self, x):
+        for module in self.module_list:
+            x = module(x)
+
+        return x
+
+if __name__ == '__main__':
+    model = Darknet53()
+    torch.save(model.state_dict(), './init_model.pt')
+    save_bin(model, 'init_model')