Model Libraries And Bbaselines

Test Environment

Important note: Due to my limited resources, only a single 1080Ti video card can be used for training, and the training cycle is long after the complete training.To illustrate this framework is trainable, inferential, and testable.The pre-training model given in this paper is only the model of 24 epoch trained, and the loss value is still declining, which has not been fully trained, so it is only for your reference.Developers with the right conditions can complete this training, and I hope you can train the good pre-training model to provide, for everyone to use.I will also make a statement and thank you in an important place.

network	pre-training data set	input size	epoch	graphics card type	inference time (FPS)	cocotools AP^val	AP₅₀	baidu network disk	Google network disk	configuration file	logs
YOLOv4	MSCOCO	608	24	2070	23ms	32.3	35.9	link Extra code:yolo	link	config file	Statistical log
YOLOv5-l	MSCOCO	640	24	2070	19ms	32.5	37.5	link Extra code:yolo	link	config file	Statistical log
PP-YOLO	MSCOCO	608	24	2070	20ms	44.5	49.4	link Extra code:yolo	link	config file	Statistical log
YOLOv4-sam	MSCOCO	608	33	2070	22ms	34.6	38.1	link Extra code:yolo	link	config file	Statistical log
YOLOv5-l+TTA	MSCOCO	640	24	2070	52ms	33.6	38.3	link Extra code:yolo	link	config file	Statistical log
YOLOv4+TTA	MSCOCO	640	24	2070	59ms	32.9	36.2	link Extra code:yolo	link	config file	Statistical log

- YOLOv4-sam：