I am new to Deep Learning and was trying to implement ResNet and VGGNet architecture on my own dataset of binary classification between roads and grass.
When i use the pretrained ResNet or VGG model they dont classify the images as road and grass
I want to learn how to manually set the output classes
CodePudding user response:
Your problem falls into one of the transfer learning approaches, which is using pre-trained model (e.g., ResNet
and VGGNet
) as a feature extractor.
This is because your dataset and the dataset used to train the pre-trained models are different. They were learned from the ImageNet dataset aimed to classify images into 1000 object categories. This means that you can't apply every part of the pre-trained model.
The CNN pre-trained model can be split into 2 main parts:
- feature extractor: groups of ConV and Pooling layer
- classifier: fully connected NN
In your case, you can use only their feature extractor part by freezing or fine-tuning them, and you require to add your own classifier to fit with your classification task.