yolov5s

[linhaoqi027]

❔Question

yolov5s is still have too many parameters for me. How should I change yolov5s.yaml to decrease parameters?(FPS is one of my evaluation.)

Additional context

[jinfagang]

@linhaoqi027 maybe yolov5nano (yolov5n) can meet your requirements.

[linhaoqi027]

where can i find it

[glenn-jocher]

@jinfagang I like it. @linhaoqi027 YOLOv5n does not exist, but you can modify the base YOLOv5 scaling constants to produce a smaller model here:
https://github.com/ultralytics/yolov5/blob/5e0b90de8f7782b3803fa2886bb824c2336358d0/models/yolov5s.yaml#L1-L5

If you reduce width_multiple by half for example, from 0.5 to 0.25, then your model will have about 1/4 the parameters, or 1.9M vs 7.5M for YOLOv5s.
python from n params module arguments 0 -1 1 1760 models.common.Focus [3, 16, 3] 1 -1 1 4672 models.common.Conv [16, 32, 3, 2] 2 -1 1 5088 models.common.BottleneckCSP [32, 32, 1] 3 -1 1 18560 models.common.Conv [32, 64, 3, 2] 4 -1 1 40640 models.common.BottleneckCSP [64, 64, 3] 5 -1 1 73984 models.common.Conv [64, 128, 3, 2] 6 -1 1 161152 models.common.BottleneckCSP [128, 128, 3] 7 -1 1 295424 models.common.Conv [128, 256, 3, 2] 8 -1 1 164608 models.common.SPP [256, 256, [5, 9, 13]] 9 -1 1 313088 models.common.BottleneckCSP [256, 256, 1, False] 10 -1 1 33024 models.common.Conv [256, 128, 1, 1] 11 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest'] 12 [-1, 6] 1 0 models.common.Concat [1] 13 -1 1 95104 models.common.BottleneckCSP [256, 128, 1, False] 14 -1 1 8320 models.common.Conv [128, 64, 1, 1] 15 -1 1 0 torch.nn.modules.upsampling.Upsample [None, 2, 'nearest'] 16 [-1, 4] 1 0 models.common.Concat [1] 17 -1 1 24000 models.common.BottleneckCSP [128, 64, 1, False] 18 -1 1 36992 models.common.Conv [64, 64, 3, 2] 19 [-1, 14] 1 0 models.common.Concat [1] 20 -1 1 78720 models.common.BottleneckCSP [128, 128, 1, False] 21 -1 1 147712 models.common.Conv [128, 128, 3, 2] 22 [-1, 10] 1 0 models.common.Concat [1] 23 -1 1 313088 models.common.BottleneckCSP [256, 256, 1, False] 24 [17, 20, 23] 1 115005 Detect [80, [[10, 13, 16, 30, 33, 23], [30, 61, 62, 45, 59, 119], [116, 90, 156, 198, 373, 326]], [64, 128, 256]] Model Summary: 191 layers, 1.93094e+06 parameters, 1.93094e+06 gradients, 4.8 GFLOPS

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.