Find Notes for the papers here
- Make it easy for anyone to understand huge libraries by taking small examples and breaking it up in that way
- Tackle all sorts of problems using pure pytorch and not relying on external libraries like Lightning/Fastai etc etc
- End up with boilerplate code for a variety of tasks
- Note that this is work in progress. And many more folders will be added in the coming days
- Look for whats new. Since most of the code remains the same, there are only a few bits that change. Find them. And youll understand.
- Search for the "#This is new" comment
- How do you find whats new? Look at the basicNet repo. This is the starter template. Everything comes from there.
- What about GANs? Look at DCGAN. It is the basic one
- Papers? Well you can refer to my notes repo. (above) or you can just read the paper (I recommend this )
-
Want to understand research papers? Check out my other repo Link
-
Comments!!
-
Standardized (mostly) code. Aka I try to be as uniform in my approach as possible
-
Might write a few blog posts
-
To be very honest, it is for my practise xD
- Can I use my own dataset? Yes of course. Just modify the dataloader. (I will be adding tutorials for that soon either way)
- No distributed training. I have only one GPU (sad) so I wont be using it
- What is the accuracy? Well run it and find out. I cant train everything to its maximum capacity due to lack of time
- Why so many files? I wonder the same sometimes. But its easier to modify for other data. And keeps my errors easily identifiable
- Why do you have so many arguments? Most of them arent even used? This is because I wanted to "standardize the codes". So I dont remove much. Just add what I need. That way its easier to see the difference between things. Just remove it if you dont want it there. (If it doesnt break anything)
All the sites I referred to for the codes here (heavily modified so you probably wont find the same things but here for future reference just in case anyone wants it). I have to admit that I relied pretty heavily on these repos. Do have a look at them if you are interested.