This an English technical report for my final project in my undergraduate days 🎓.
Also, this project might be my last computer vision one. So I decided to build this repo and share my interesting findings. 🌟
The technical report is in tech report, and the latex project is provided.
Why $21 ? That means the biggest model YOLO11x-Earth is pretrained on xView on a RTX 4090 for less than 70 hours.
The price of renting a RTX 4090 is $0.3/h on AudoDL.
At first, I want to publish it on Arxiv. But it's a pity that I do not have the endorsement required.
It's framework:
It's performance after finetuning on DIOR:
I provide the model config file yolo11-earth.yaml in model and some useful toolkits in utils , like how to split xView and how to build a gradio app for your model.
There is no furthur plan to share my training scripts. Becasue with the model yaml you can easily reproduce my work with
I'm glad to share my weights.
Now I'm looking for a efficient methods to share these weights (about 50 files), maybe Google Drive.
Let me take a break at first plz. 🎶