Transfer learning & pretrained model | Normalization

目录

Transfer learning & pretrained model training tps

文章介绍了transfer learning的用途以及实现（ie. 使用pretrained model）；

对CNN的理解：Convolutional base(The main goal of the convolutional base is to generate features from the image.) (以特征工程处理的角度看conv base)+ Classifier；

重利用pretrained model的三套fine-tuning策略：

Transfer learning process：

Classifiers选择

Normalization(Min-max norm & z-score norm)

My baseline model: nnU-Net: Self-adapting Frameworkfor U-Net-Based Medical Image Segmentation

---

Transfer learning & pretrained model training tps

文章介绍了transfer learning的用途以及实现（ie. 使用pretrained model）；

对CNN的理解：Convolutional base(The main goal of the convolutional base is to generate features from the image.) (以特征工程处理的角度看conv base)+ Classifier；

重利用pretrained model的三套fine-tuning策略：

（其中S1 需要从头完全train，因此需要大量的data；S2 这种情况往往是数据量 vs Frozen程度。由于lower layers对应general features，所以lower layers训好的参数可以使用，frozen的程度考虑到overfitting的可能其取决于数据量的大小；S3 对应数据量很小或者原始model的任务与当前任务有很大相似的情况；另外在fine-tuning S1和S2时要careful）

Transfer learning process：

1、首先选择一套pretrained model，keras版本可用；

2、搞清当前问题情况：

（关于数据集大小和为相似的经验之谈：As a rule of thumb, consider that your dataset is small if it has less than 1000 images per class. Regarding dataset similarity, let common sense prevail. For example, if your task is to identify cats and dogs, ImageNet would be a similar dataset because it has images of cats and dogs. However, if your task is to identify cancer cells, ImageNet can’t be considered a similar dataset.）

3、Fine-tune your model.

（与2相对应，选择合适的Frozen程度）

Classifiers选择

---

Normalization(Min-max norm & z-score norm)

min-max norm：

效果图（问题：对outliers的敏感）：

z-score norm：

效果图：

总结：

• Min-max normalization: Guarantees all features will have the exact same scale but does not handle outliers well.
• Z-score normalization: Handles outliers, but does not produce normalized data with the exact same scale.

---

My baseline model: nnU-Net: Self-adapting Frameworkfor U-Net-Based Medical Image Segmentation

Fabian Isense（Division of Medical Image Computing, German Cancer Research Center (DKFZ),Heidelberg, Germany）