When my training loss goes down, my evaluation loss goes up. What is this a symptom of?

Question

I'm training a multi-label image classifier with a dataset of 4096 images and a network model based off of DenseNet, using an SGD optimizer. The pink graph has a learning rate of 0.01, and the blue graph has a learning rate of 0.001. Both have a momentum of 0.9. What is this a symptom of? Do I need more data?