python / imaging project - data manipulation and image filtering

background: I have ~100k images that I am doing deep learning on -images all stored in cloud, coded in python / keras I have built pipeline, but need assistance with a few steps: task list: -most efficient way of reviewing all images to filter out images of poor quality most efficient way to view a subset of images based on a set of conditions from a dataframe -converting images to numpy arrays or, preferably HDF5 databases -using iotop to monitor GPU usage please respond with cost and time estimate of undertaking this task, or any other further questions