docs(r-cnn): 添加训练结果分析

object-detection-algorithm · Apr 3, 2020 · 0a0dc12 · 0a0dc12
1 parent a89dbf2
commit 0a0dc12
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 2 deletions.
diff --git a/docs/index.md b/docs/index.md
@@ -13,6 +13,8 @@
 3. 分类器训练
 4. 目标检测器实现
 
+**本仓库最终实现一个汽车类别目标检测器**
+
 ## 模块构成
 
 1. 区域建议生成：`selectivesearch`算法实现，生成类别独立的区域建议

diff --git a/docs/分类器训练.md b/docs/分类器训练.md
@@ -128,4 +128,6 @@ val Loss: 1.0424 Acc: 0.9246
 remiam negative size: 365403, acc: 0.9981
 Training complete in 55m 50s
 Best val Acc: 0.924645
-```
+```
+
+**经过`2`轮训练后，分类器损失就开始收敛**
diff --git a/docs/微调模型.md b/docs/微调模型.md
@@ -156,4 +156,10 @@ val Loss: 0.5996 Acc: 0.8718
 
 Training complete in 461m 14s
 Best val Acc: 0.873228
-```
+```
+
+模型在训练过程中呈现过拟合现象，可考虑调整学习率、添加权重衰减以及更换优化器的方式：
+
+1. 学习率从`1e-3`调整为`1e-4`
+2. 添加`L2`权重衰减，衰减因子为`1e-4`
+3. 使用`Adam`替换`SGD`