Skip to content

Commit

Permalink
Merge pull request #115 from insurgent92/issue_40
Browse files Browse the repository at this point in the history
#40: Translate 5.3 - paragraph 8.2 / 11
  • Loading branch information
visionNoob authored Oct 2, 2018
2 parents d7a3d32 + 1a09316 commit 3e41c6c
Showing 1 changed file with 18 additions and 0 deletions.
Original file line number Diff line number Diff line change
Expand Up @@ -1063,6 +1063,24 @@
"As a reminder, this is what our convolutional base looks like:"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Fine-tune을 하기 위해서는 우선 VGG16의 모든 convolution base를 freeze하고, 새롭게 추가시켜서 램덤으로 초기화된 분류기를 학습시킬 필요가 있습니다. 새롭게 추가시킨 분류기가 학습이 완료된 상태에서만 convolution base를 fine-tune 할 수 있는 것이죠. 만약 분류기가 학습이 되어있는 상태가 아니라면 학습 도중 전달되는 에러신호가 아주 클 것이고, 앞서 사전에 학습한 특징들이 전부 손상될 것입니다. 네트워크를 fine-tune하는 방법은 다음과 같습니다 : \n",
"\n",
"* 1) Add your custom network on top of an already trained base network.\n",
"* 2) Freeze the base network.\n",
"* 3) Train the part you added.\n",
"* 4) Unfreeze some layers in the base network.\n",
"* 5) Jointly train both these layers and the part you added.\n",
"\n",
"We have already completed the first 3 steps when doing feature extraction. Let's proceed with the 4th step: we will unfreeze our `conv_base`, \n",
"and then freeze individual layers inside of it.\n",
"\n",
"As a reminder, this is what our convolutional base looks like:"
]
},
{
"cell_type": "code",
"execution_count": 17,
Expand Down

0 comments on commit 3e41c6c

Please sign in to comment.