Based on YOLOv8l, we trained it with SimCLR - Powerful Self-Supervised Pretraining Method.
There are three versions of model backbones:
- YOLOv8l pretrained from scratch with ~500 images
- YOLOv8l pretrained from scratch with ~1000 images
- COCO-Pretrained YOLOv8l finetuned with SimCLR with ~1000 images
The images are filmed ourselves, which is Korean Subway (Seoul Station to Seoul National University)
