So, I stumbled upon this thing called “yunli hsr kit” the other day, and I was like, “What the heck is that?” I’m always up for trying new techy stuff, so I decided to give it a whirl. I’m gonna walk you through what I did, no fancy tech jargon, just plain English.
Getting Started
First things first, I needed to, you know, actually get my hands on this “kit”. I did a quick search but didn’t find that much helpful info, Seems nobody shared it before. I figured I had to built everything myself and there’s no place I can download the complete kit.
The Deep Dive
I decided to build everthing from the ground up.
install everthing needed:
- create a new conda env
- install pytorch
- install other package,just use pip install, it is easy.
Then, download the offical code:
- clone it.
Prepare the pretrain weights:
- download it, I am not going to share where I got it from, just search it.
- put it in a safe place.
Prepare the dataset:
- They have lots of it, all is open source, that is great.
- I only tried a small one, it works, that’s good enough.
- Follow the official document to process the dataset.
Finetune
I spent a few hours to fine tune everything, following the instrcution.
And I opened the tensorboard to check the loss, it keeps going down!
My Two Cents
Okay, so after all that tinkering, is it difficult to build everything? It’s not that hard, just follow the steps, and be patient. I successfully run everything. I believe you can do it too.
