From 756c4c2e184116b96391554a8f911eaa36f76550 Mon Sep 17 00:00:00 2001
From: Bubbliiiing <47347516+bubbliiiing@users.noreply.github.com>
Date: Sat, 11 Jul 2020 17:13:05 +0800
Subject: [PATCH] Update train.py

---
 train.py | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/train.py b/train.py
index cd5fdef..da053df 100644
--- a/train.py
+++ b/train.py
@@ -83,6 +83,10 @@ def fit_one_epoch(net,focal_loss,epoch,epoch_size,epoch_size_val,gen,genval,Epoc
     torch.save(model.state_dict(), 'logs/Epoch%d-Total_Loss%.4f-Val_Loss%.4f.pth'%((epoch+1),total_loss/(epoch_size+1),val_loss/(epoch_size_val+1)))
     return val_loss/(epoch_size_val+1)
     
+#----------------------------------------------------#
+#   检测精度mAP和pr曲线计算参考视频
+#   https://www.bilibili.com/video/BV1zE411u7Vw
+#----------------------------------------------------#
 if __name__ == "__main__":
     #-------------------------------------------#
     #   训练前，请指定好phi和model_path
@@ -106,6 +110,9 @@ if __name__ == "__main__":
     # 创建模型
     model = EfficientDetBackbone(num_classes,phi)
 
+    #------------------------------------------------------#
+    #   权值文件请看README，百度网盘下载
+    #------------------------------------------------------#
     model_path = "model_data/efficientdet-d0.pth"
     # 加快模型训练的效率
     print('Loading weights into state dict...')
@@ -135,6 +142,14 @@ if __name__ == "__main__":
     num_val = int(len(lines)*val_split)
     num_train = len(lines) - num_val
     
+    #------------------------------------------------------#
+    #   主干特征提取网络特征通用，冻结训练可以加快训练速度
+    #   也可以在训练初期防止权值被破坏。
+    #   Init_Epoch为起始世代
+    #   Freeze_Epoch为冻结训练的世代
+    #   Epoch总训练世代
+    #   提示OOM或者显存不足请调小Batch_size
+    #------------------------------------------------------#
     if True:
         #--------------------------------------------#
         #   BATCH_SIZE不要太小，不然训练效果很差
@@ -208,4 +223,4 @@ if __name__ == "__main__":
 
         for epoch in range(Freeze_Epoch,Unfreeze_Epoch):
             val_loss = fit_one_epoch(net,efficient_loss,epoch,epoch_size,epoch_size_val,gen,gen_val,Unfreeze_Epoch,Cuda)
-            lr_scheduler.step(val_loss)
\ No newline at end of file
+            lr_scheduler.step(val_loss)
-- 
GitLab