search

PINTO0309 / MobileNetv2-SSDLite

My proprietary procedure. Caffe implementation of SSD and SSDLite detection on MobileNetv2, converted from tensorflow.
https://github.com/chuanqi305/MobileNetv2-SSDLite
MIT License
23 stars 11 forks source link

the mobilessdlite coco model is not converted correct? #3

Closed ChenjingYu1993 closed 5 years ago

ChenjingYu1993 commented 5 years ago

the mobilessdlite coco model is not converted correct? when i convert the model to ncnn, the result is not correct~

PINTO0309 commented 5 years ago

@1976277169

Yes. It's exactly as you say. I understand. Please see #2. Please introduce custom Caffe corresponding to Relu6.

ChenjingYu1993 commented 5 years ago

thx for your patient answers, a new question is: if i want to convert the coco mobilessd model to ncnn correctly, what should i do, can you give me some advice?

PINTO0309 commented 5 years ago

@1976277169 That's an interesting library! I have not tried it yet, so I can not give you the proper advice. I think I will try it when I have a lot of free time.

PINTO0309 commented 5 years ago

@1976277169 btw, I intentionally leave "deploy.prototxt" destroyed. Please refer to the following, first replace the "ReLU" layer with "ReLU6". https://github.com/PINTO0309/MobileNetv2-SSDLite#note

ChenjingYu1993 commented 5 years ago

Thank you for your generous help! I'll have a try, Thank you , Thank you, Thank you~~~~

PINTO0309 commented 5 years ago

I have organized the procedure correctly. https://qiita.com/PINTO/items/fe98bde40d0ffd13455e