could provide the yolov7_SiLU pretrained model，I test the yolov7 model, so slowly than yolov5s yolov6s?

WongKinYiu / yolov7

Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

GNU General Public License v3.0

13.02k stars 4.12k forks source link

could provide the yolov7_SiLU pretrained model，I test the yolov7 model, so slowly than yolov5s yolov6s? #16

Open xinsuinizhuan opened 2 years ago

xinsuinizhuan commented 2 years ago

windows vs 2019 cuda 11.3 cudnn 8.2.3 tensorrt 8.2 3060ti amd 5900H

openvino forward: yolov5s: 90ms yolov6s 75, yolov7 282ms opencv dnn forward: yolov5s:21, yolov6s 18ms, yolov7 25ms tensorrt yolov5s: 6ms yolov6s 5ms yolov7 11ms

WongKinYiu commented 2 years ago

yolov7-tiny

xinsuinizhuan commented 2 years ago

yolov7-tiny

is there the yolov7-tiny pretrained model?

WongKinYiu commented 2 years ago

yolov7-tiny.weights

xinsuinizhuan commented 2 years ago

yolov7-tiny.weights

could provide the yolov7-tiny.pt pretrained model?

WongKinYiu commented 2 years ago

OK, I will generate one soon.

Aruen24 commented 2 years ago

can we use the cfg/deploy/yolov7-tiny-silu.yaml to train pytorch?

WongKinYiu commented 2 years ago

yes, but silu is not good for edge device, we do not recommend use silu in tiny model.

Aruen24 commented 2 years ago

yes, but silu is not good for edge device, we do not recommend use silu in tiny model. can support the yolov7-tiny.yaml not silu for pytorch.

WongKinYiu commented 2 years ago

@xinsuinizhuan yolov7-tiny.pt

xinsuinizhuan commented 2 years ago

@xinsuinizhuan yolov7-tiny.pt

OK.Thank you. Let me have a try. VS yolov5s and yolov6s

Aruen24 commented 2 years ago

the yolov7-tiny.pt include the silu?

WongKinYiu commented 2 years ago

No, it uses Leaky ReLU.

Aruen24 commented 2 years ago

the cfg/deploy/yolov7-tiny.yaml file can release?

wilile26811249 commented 2 years ago

@wang911205 yolov7-tiny.yaml

Aruen24 commented 2 years ago

the file is yolov7-tiny-silu.yaml include the silu

atangfan commented 2 years ago

Do you have pretrained YOLOv7 with LeakyReLU? How many mAP drop comparing to SILU version?

JNH-LD commented 2 years ago

ubuntu cuda 11.2 cudnn 8.1.1 1080ti

pt inference only: yolov5n: 5.5ms yolov7-tiny: 7. 5ms

xinsuinizhuan commented 2 years ago

windows vs 2019 cuda 11.3 cudnn 8.2.3 tensorrt 8.2 3060ti amd 5900H

tensorrt: yolov7-tiny: 3ms

WongKinYiu commented 2 years ago

Could you help for testing size 512, 416, and 320, thanks.

serser commented 1 year ago

So no silu version is provided yet? The paper is reporting a 38.7% mAP (can this be released?), but I tested with the provided 'yolov7-tiny.pt' (actually leaky-relu version), got only 37.4%.