When I set the same data and environment and run test.py, why is yolov3 faster than yolov5?

[LEEGILJUN]

❔Question

The basic hyper parameters and environment were not changed. and same image
batch-size : 1
image size : 800x800
image : gray scaled Satellite image
Number of objects : 7
pretrained : coco2017 by yolov3 and v5
code : ultralytics pytorch code for v3 and v5

result
yolov3 : 40.7/1.4/42.2 ms inference/NMS/total
yolov5 : 71.8/1.2/73.0 ms inference/NMS/total

mAP
yolov3 : 0.707, 0.436
yolov5 : 0.785, 0.552

Additional context

When I studied the yolo series, I think that v5 should be faster than v3 not only in performance but also in inference speed. And I did several experiments by k-folding the data currently in use. But the result was always yolov3 faster. Other kinds of data were the same when I did the experiment. I would like to know why this is happening.

[glenn-jocher]

@LEEGILJUN that's not correct. Our YOLOv3 README table has a direct comparison between YOLOv3 and YOLOv5, and YOLOv5 exceeds YOLOv3 in every metric, including speed, size, FLOPS, parameters, mAP, etc.
https://github.com/ultralytics/yolov3#pretrained-checkpoints

Pretrained Checkpoints

Model	size (pixels)	mAPval 0.5:0.95	mAPtest 0.5:0.95	mAPval 0.5	Speed V100 (ms)		params (M)	FLOPS 640 (B)
YOLOv3-tiny	640	17.6	17.6	34.8	1.2		8.8	13.2
YOLOv3	640	43.3	43.3	63.0	4.1		61.9	156.3
YOLOv3-SPP	640	44.3	44.3	64.6	4.1		63.0	157.1
YOLOv5l	640	48.2	48.2	66.9	3.7		47.0	115.4