This project is inspired by cs230 project.
We analyze 2 chest datasets from kaggle:
- Chest X-Ray Images (Pneumonia) - contains around 5K examples (pneumonia set); and
- NIH Chest X-rays - contains around 100K examples (NIH dataset);
That's a typical image classification problem with 2 or 3 classes for the pneumonia set and 15 classes for the NIH dataset. We are going (preliminary):
- to use transfer learning (probably
resnet50) on the pneumonia set with some augmentation; - retrain
resnet50(some convolution layers) on the NIH dataset and use it for transfer learning on the pneumonia set (challenge);
We're going to use tf.keras from tensorflow 2.0. We train in Google Cloud:
- Machine type:
n1-standard-8 (8 vCPUs, 30 GB memory); - GPUs:
1 x NVIDIA Tesla T4;