How to achieve distributed deep learning framework for computer vision?

I am Ph.D. student in Economics, and currently working on project of deep learning-based object detection from satellite imagery.
I am struggling to get an idea to conduct distributed deep learning on my current task. I have access to PySpark 2.4.0 and PyTorch 1.9.0 at my institute. I was wondering if there is useful resources or tutorials that can help me apply these tools to my work.
Specifically, I’m interested in learning about the common approaches used in distributed deep learning frameworks for image processing or computer vision tasks, especially when it comes to working with Pytorch.
Plus, I also would like to know whether PySpark is good or common method to load image data from HDFS for building object detection models, and when I conver Pyspark.dataframe data format to proper data format for PyTorch, does distributed or parallelized nature remain?

The DistributedDataParallel tutorial might be a good starter as it would walk you through the usage of DDP and should be relatively easy to apply for any model. You could then check more advanced techniques of model sharding if needed.