Table 4.

Comparison of the number of parameters and flops computational complexity of models. Input size: (600,400).

Method	Params/M	FLOPs/GFLOPs
Two-stage anchor-based
Faster R-CNN	41.12	61.52
Mask R-CNN	43.75	113.0
MS R-CNN	60.01	113.0
One-stage anchor-based
RetinaNet	36.10	50.55
YOLOv3	61.52	47.88
ATSS	31.89	49.79
One-stage anchor-free
FCOS	31.84	48.62
FoveaBox	36.01	50.06
Ours
FCOS-LSC	38.65	38.72