yolov8

Zone Evaluation on YOLOv8

Download YOLOv8 from https://github.com/ultralytics/ultralytics and install the dependencies.

Make sure you have compiled our pycocotools.

Replace ./yourpath/ultralytics/ultralytics/yolo/v8/detect/val.py with our provided file val.py.

You have to install numpy-1.23.3 to avoid error.

conda install numpy=1.23.3

Recompile

 pip install -e '.[dev]'

Then run,

yolo mode=val task=detect data=coco.yaml model=yolov8n.pt device=\'0,1\'

And you will see the zone evaluation results (YOLOv8-n):

Zone:, ZP, ZP50, ZP75, ZPs, ZPm, ZPl, ZR1, ZR10, ZR100, ZRs, ZRm, ZRl
z05:  [       37.3        52.5        40.5        18.5          41        53.5          32        53.2        58.8        36.6        65.3        76.8]
---------------------------------------
z01:  [       26.4        37.5        28.6        15.7        34.4        45.2        33.8        48.8        50.8        35.3        62.3        69.9]
z12:  [         34        47.9        36.6        17.4        41.1        48.4        38.7        54.7        56.9        34.9        62.7        73.9]
z23:  [         37        51.2        39.8        18.2        41.8        52.9        38.2        55.1        57.2          37        64.6        72.3]
z34:  [       35.8        50.5        38.4        19.8        41.5        52.5        39.2          53        54.9        34.4        63.4        70.3]
z45:  [       39.6        54.7        43.3        24.6          43        54.5        48.3        57.9        58.7        38.1        63.9        76.8]
---------------------------------------
ZP_variance:  [     19.717      34.237       24.03      9.1978      9.3539      11.555      22.441      9.0023      7.3606      1.9472     0.68575       6.403]
SP, SP50, SP75, SPs, SPm, SPl, SR1, SR10, SR100, SRs, SRm, SRl
SP: [       32.3        45.4        34.8        17.5        38.9        48.9        37.3        52.6        54.6        35.5          63        71.8]

Test resolution: 640.

Detector	SP	$\text{ZP}^{0,5}$	Variance	$\text{ZP}^{0,1}$	$\text{ZP}^{1,2}$	$\text{ZP}^{2,3}$	$\text{ZP}^{3,4}$	$\text{ZP}^{4,5}$	params (M)	FLOPs (B)
YOLOv5-s	33.3	37.4	10.5	28.8	34.9	36.9	35.1	38.4	7.2	16.5
YOLOv8-n	32.3	37.3	19.7	26.4	34.0	37.0	35.8	39.6	3.2	8.7
YOLOv5-m	40.8	45.2	12.9	36.0	42.3	44.5	43.2	46.7	21.2	49.0
YOLOv8-s	39.8	44.9	24.4	33.4	42.2	44.3	43.2	48.5	11.2	28.6

Discussion: If we compare YOLOv8 and YOLOv5 with similar AP, the improvement of YOLOv8 mainly comes from large objects and central zone. Besides, YOLOv5 performs better in spatial equilibrium (lower variance).