dog-qiuqiu / mobilenet-yolo Goto Github PK
View Code? Open in Web Editor NEWMobileNetV2-YoloV3-Nano: 0.5BFlops 3MB HUAWEI P40: 6ms/img, YoloFace-500k:0.1Bflops 420KB:fire::fire::fire:
License: Other
MobileNetV2-YoloV3-Nano: 0.5BFlops 3MB HUAWEI P40: 6ms/img, YoloFace-500k:0.1Bflops 420KB:fire::fire::fire:
License: Other
请问下大佬, 如何训练 Ultra lightweight 106 point face-landmark model ?
非常感谢。
大大,可不可说明一下如何训练及使用,对于新手来说很有必要,谢谢!!
想尝试魔改下模型,重新训练,但不清楚yoloface的数据集是哪个
你好,想了解一下yolov4或者yolov4-tiny目前能在mnn推理框架中使用吗,有没有推理时间的测试数据啊
大佬,有什么建议将yolov-4转为caffe吗,现在卡到了通道数减半这个操作上,caffe有哪个对应层可以对应吗,或者几个层组合实现这个功能,非常感谢
请问一下我们我的电脑运行yolov4-tiny 能跑80fps,但是MobileNetV2-YOLOv3-Lite只能跑11fps,这是怎么回事呢?
armv8.2变快啦~~
模型是YOLO fastest的模型,转换完caffe模型后,使用detector.py进行测试的时候,报错KeyError: 'layer79-conv',这个地方需要修改key吗
因为CUDA Version: 11.0,所以装的cudnn为 v8.0.4,设置cuda=1,cudnn=1,然后编译,然后报错了,如下
error:CUDNN_CONVOLUTION_FWD_PREFER_FASTEST undeclared (first use in this function); did you mean CUDNN_CONVOLUTION_BWD_FILTER_ALGO_3? int forward_algo = CUDNN_CONVOLUTION_FWD_PREFER_FASTEST; ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ CUDNN_CONVOLUTION_BWD_FILTER_ALGO_3 compilation terminated due to -Wfatal-errors.
网上有找原因,可能是cudnn版本过高导致,只能cudnnv3-v7版本,但是对应cuda11.0的cudnn只有v8版本的
然后同样的配置编译AlexeyAB/darknet的版本能够通过,但我想做mobilenet版本的yolo
which dataset is trained for MobileNetV2--Lite.conv.57 ?
COCO? or VOC?
請問能提供 MNN 的 sample code嗎?
Hi.
Very interested to yoloface-500k model.
if input image size is 352*228, which size of face I can detect at lease.
Please notice me smallest face size could be detected in yoloface-500k-v2.
Thanks.
感谢分享,想了解下weights是如何转换到mnn model,有参考的链接分享下嚒,是weights->onnx->mnn流程嚒?
Hello,
Thanks for the great repo.
I was interested in the app so I downloaded & ran on my android device, and the app was closed right after I opened it.
I downloaded via link below.
https://github.com/dog-qiuqiu/Android_MobileNetV2-YOLOV3-Nano-NCNN/blob/master/app/release/MobileNetv2-yolov3-nano.apk
is there any limitation or options for running app?
Thanks a lot.
caffemodel or darknet model ,How to convert yoloface-500k model to tensorrt
貌似yolov3-nano训练出的模型caffe输出正常,但是mnn输出就是乱的,我对比了两个模型每一层输出,发现是在layer72-upsample层开始不一样,感觉是
#layer {
# bottom: "layer71-route"
# top: "layer72-upsample"
# name: "layer72-upsample"
# type: "Upsample"
# upsample_param {
# scale: 2
# }
#}
layer {
bottom: "layer71-route"
top: "layer72-upsample"
name: "layer72-upsample"
type: "Interp"
interp_param {
height:20 #upsample h size
width:20 #upsample w size
}
}
模型转换上有点瑕疵,请问大大怎么修改才能解决?谢谢!!
README
@dog-qiuqiu Hi! Thank you for your work. Please clarify why batch is so small (16) comparing to default yolov3 (64):
https://github.com/dog-qiuqiu/MobileNetv2-YOLOV3/blob/2cda47a036e0378cb6c988326c3b53016eeb2b4f/MobileNetV2-YOLOv3-SPP/MobileNetV2-YOLOv3-SPP.cfg#L2
在手机上验证了这个效果相当不错,想把它移植到海思的nnie上。我自己训练出来的模型,经过ruyi 转换没有啥异常,放到板子上跑的时候检测结果不准确,或者说完全错误,没有一个目标对得上。不知道那里除了问题,网上查了一圈有说图片预处理不对,我目前转模型的预处理就是 图片 RGB,还有一个就是 data_scale 1/256 ,不知还有没有什么其他的地方没弄对?? @dog-qiuqiu 希望能指点一下,谢谢!!!!
请问下,MobilenetV2-Yolov3-lite如何转换到ncnn下?yolo-fastest如何转化?我训练的数据要么转化成功加载的时候直接崩溃,要么就是无法正常转换。
你好,如果要用MobileNetV2-YOLOv3-Lite训练自己的数据集,直接用你提供的cfg和weight文件和MobileNetV2--Lite.conv.57,在darknet下就可以了吗?
Hello!
I've met a bug when I use your resnext152-32x4d.cfg file. My project was forked from ultralytics/yolov3.
Bug description: The size of feature maps are unmatched to do 'shortcut' op.
Where: The config file of ResneXt152 backbone - resnext152-32x4d.cfg.
The code as follow is one part of the cfg file and it's shown that feature maps in sample level X/2 cannot ADD with sample level X.
[shortcut] # -4 (sample level X)
...
[convolutional]# -3
...
[convolutional]# -2 (Downsample)
groups = 32
batch_normalize=1
filters=128
size=3
stride=2
pad=1
activation=leaky
[convolutional]# -1 (input of the shortcut)
...
[shortcut]# 0 (sample level X/2)
from=-4
activation=leaky
python darknet2caffe.py MobileNetV2-YOLOv3-Nano-voc.cfg MobileNetV2-YOLOv3-Nano-voc.weights MobileNetV2-YOLOv3-Nano-voc.prototxt MobileNetV2-YOLOv3-Nano-voc.caffemodel
我想问问VOC的平均精度是用的VOC2007的测试集,那训练集是VOC07+12还是同时在COCO上也训练了呀?谢谢
OS : ubuntu18.04
Follow the instructions, install and build the caffe, able to convert nano, lite, the error messages are
Traceback (most recent call last):
File "darknet2caffe.py", line 443, in <module>
darknet2caffe(cfgfile, weightfile, protofile, caffemodel)
File "darknet2caffe.py", line 8, in darknet2caffe
net_info = cfg2prototxt(cfgfile)
File "darknet2caffe.py", line 309, in cfg2prototxt
bottom2 = topnames[prev_layer_id2]
KeyError: 126
I hope to train yoloface-500k model on custom database.
How to train this model?
Where can I get trainer for this model.
使用了最新的NCNN,但项目里面MobileNetV2-YOLOv3-Nano-voc模型还是无法转化,报错:
(py35) qglobal@qglobal-ThinkStation-P328:~/ncnn/build/tools/caffe$ ./caffe2ncnn MobileNetV2-YOLOv3-Nano-voc.prototxt MobileNetV2-YOLOv3-Nano-voc.caffemodel
[libprotobuf ERROR google/protobuf/text_format.cc:274] Error parsing text-format caffe.NetParameter: 2066:20: Message type "caffe.LayerParameter" has no field named "upsample_param".
read_proto_from_text failed
大佬,你这个表里模型测试的NCNN模型是咋转化的
Network | VOC mAP(0.5) | COCO mAP(0.5) | Resolution | Inference time (NCNN/Kirin 990) | Inference time (MNN arm82/Kirin 990) | FLOPS | Weight size |
---|---|---|---|---|---|---|---|
MobileNetV2-YOLOv3-Lite | 72.61 | 36.57 | 320 | 31.58 ms | 18 ms | 1.8BFlops | 8.0MB |
MobileNetV2-YOLOv3-Nano | 65.27 | 30.13 | 320 | 13 ms | 5 ms | 0.5BFlops | 3.0MB |
YOLOv3-Tiny-Prn | & | 33.1 | 416 | 36.6 ms | & ms | 3.5BFlops | 18.8MB |
YOLO-Nano | 69.1 | & | 416 | & ms | & ms | 4.57BFlops | 4.0MB |
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.