Hi, thanks for sharing your implementation! This is awesome work and helps others to catch up easily. I am trying to play with your code and meet a few questions. It will be great if you could spare some time to reply.
You mentioned the "ours" method which achieves the best performance. Would you mind to share what changes have you made compared with MRNA?
You've provided two pre-trained inception nets for the frontal and lateral images. I am wondering about which task have you fine-tuned the inception net to obtain these two pre-trained models? Is it fine for us to load the original inception nets for both images and train from scratch?
Are you training on one GPU?
How do you separate the Normal Reports and Abnormal Reports?
Hi, thanks for sharing your implementation! This is awesome work and helps others to catch up easily. I am trying to play with your code and meet a few questions. It will be great if you could spare some time to reply.
Looking forward to hearing from you :-)