Transfer Learning | Deep Learning

Live Engine

Select Topic

easyTransfer Learning

A team fine-tunes a ResNet-50 pretrained on ImageNet for classifying satellite images. They use feature extraction (freeze all layers, train only the final classifier). After 20 epochs, validation accuracy plateaus at 62%. The same team fine-tunes all layers and achieves 84%. What does this reveal about the feature extraction vs fine-tuning decision?

Live Engine

Select Topic

easyTransfer Learning

A team fine-tunes a ResNet-50 pretrained on ImageNet for classifying satellite images. They use feature extraction (freeze all layers, train only the final classifier). After 20 epochs, validation accuracy plateaus at 62%. The same team fine-tunes all layers and achieves 84%. What does this reveal about the feature extraction vs fine-tuning decision?