Hi Pytorch forum! Recently I came up with an idea for my side project which is using depth estimation to build a mini car toys that can keep itself away from obstacles but I don’t know what’s a proper approach to solve real world problem in this field? Is that an ideal way to just clone the best method out there and fine tune to fit this problem or read through paper proposed method and write and train model from scratch?
Hey there,
Well, it depends on what you want to focus on first, also, how much experience do you have in reading the papers and creating neural networks. If your focus is to make the mini car work and perform obstacle avoidance, then, it´s a lot better for you to clone a project and fine tune it to your needs, since it will provide to you a lot quicker proof of concept than starting from scratch. I would only suggest to read proposed methods and train it from scratch only when you are very experienced in this and know what you want to try to accomplish.
Also, another point why cloning and fine tune it first is better, is due to the fact that when you further develop a project, you discover new needs or deprecate others. Software is always evolving and changing. For instance, you may implement from scratch a proposed method you found, but, later on, you discover that maybe the network is not precise enough or its too big to fit your processing needs! Then what happens there? You learned how to do it, sure, but you took a not so much time efficient path(time is also a precious resource 
)
Fast.ai touches the subject of how to learn new subjects, if you are interested. If you are new / semi - new, its better to fine tune networks that are already implemented and tested and, later, touching the “lower level” when you need something more.
A summary: 99% of the times is better to grab a working network and fine tune it. It will provide you a quicker proof of concept, you will learn many requirements on the way and change things as needed, more rapidly and time efficient.