This is an AIA final assignment. Our team members are: YirenQIU and Ghous; we chose task 2 --- Create your own object detection algorithm, and compare the results w.r.t. a pre-trained YOLO model.
We chose to modify the neck part of RT-DETR and replace it with the GSConv module to reduce the amount of calculation and thus reduce the complexity of the model.
We provide the following reports:
- A presentation PPT file, currently only lacking experimental comparison and analysis(We hope this could be helpful for beginners understanding the use of ultralytics' source codes easier, about where are the source codes exist, why the configuration looks like that and how to format, where and how to register your design's extra_module...)
- The source code of the modified model --- GSconv-rtdetr based on Ultralytics platform 2024.11.28
- Data report on the modified model --- GSconv-rtdetr on visdrone 2024.11.28
The repository will keep in updating in the future.
2024-12-19： During the defense week, I learnt a lot of technology about making a awesome slide, so I remake a great interesting slides for the happy presentation, this slides created by canvas, please read that online to get the best reading experience.