hjimce / darknet_mobilenet

mobilenet 与darknet yolo
Other
94 stars 52 forks source link

darknet_mobilenet pre-trained model #2

Open kingvision opened 6 years ago

kingvision commented 6 years ago

Hi: can you provide a darknet_mobilenet pre-trained model?

thanks

TaihuLight commented 6 years ago

https://github.com/zunzhumu/darknet-mobilenet

kingvision commented 6 years ago

@TaihuLight Thanks, i use the model, but i detect no any objects. where i should modify the code? thanks

wuxiaomin0110 commented 6 years ago

@kingvision I detect no object, either. Did you find a solution? Thanks.

wuxiaomin0110 commented 6 years ago

@hjimce Hi: Is "mobilenet.weights" the pre-trained model corresponding to cfg/mobilenet_imagenet.cfg? I use the model, but don't detect anything. How do I adjust the code? Thank you.

hjimce commented 6 years ago

@vmaxxwuxiaomin mobilenet_imagenet.cfg and pre-trained model mobilenet.weights are train for imagenet classification

hjimce commented 6 years ago

@vmaxxwuxiaomin no object detection

wuxiaomin0110 commented 6 years ago

@hjimce OK. Thanks so much. Classification by your model is OK. It's amazing! I wonder if I wanna detect objects, I need to modify the .cfg file and re-train the .weights file, right?

hjimce commented 6 years ago

@vmaxxwuxiaomin yes,you can modify the mobilenet_imagenet.cfg like tiny-yolo.cfg ,except the convolution layer

wuxiaomin0110 commented 6 years ago

@hjimce OK. I will try. Thanks so much.

Li-Lai commented 6 years ago

@hjimce Thank you for sharing the code. I want ro know the mobilenet.weights you provide is match with mobilenet_imagenet.cfg? I can run classification command, but can not get right classification results.

hjimce commented 6 years ago

@CBIR-LL sorry,I forget it。you can train it by youself on imagenet-1k data

Li-Lai commented 6 years ago

@vmaxxwuxiaomin @hjimce Did you try to detect objects using mobilenet network structure ? Is there any improvement in speed ?

qxin commented 6 years ago

@vmaxxwuxiaomin hi, how to spell the classification command. thanks ~ (I found it ) it workswell, still thanks. ## ./darknet classify cfg/mobilenet-1k.cfg mobilenet.weights data/dog.jpg

ljh14 commented 6 years ago

When I train the model,why does the loss function stop drop down at about 23 ...It is so unbelievable...It makes me so frustrated

A-levy commented 6 years ago

@wuxiaomin0110 我最近也在搞这个mobilenet的检测,请问您实现了吗?可以交流一下吗?谢谢!

DW1HH commented 5 years ago

@A-levy 你QQ多少

aimhabo commented 5 years ago

请问这样的训练情况正常么?感觉loss变成nan有点不对劲 Is this training normal? Feeling loss becomes a bit wrong with nan screenshot from 2018-12-26 17-55-50