This is my final project of Deep Learning on Computer Vision course. In fact, it is a work on few-shot learning.
We implemented a Siamese Network to compute the similarit score between images. However, we found it difficult to directly train a Siemese Network on the whole dataset. Thus, we designed a training schedule that gradually increases the training difficulty. After it converges, we fine-tuned the network on the images which has few labels. Through this special training schedule, we pass all the requirements in this task.