靓仔我又来了

[ZHEQIUSHUI]

我在xavier上试了 yolo是可以的 但是openpose没有结果，所以我又在我的电脑上试了，openpose还是没结果,但是yolo也是一样可以的。难道我用错模型了...

/testopenpose --prototxt ~/lib/openpose/openpose/models/pose/body_25/pose_deploy.prototxt --caffemodel ~/lib/openpose/openpose/models/pose/body_25/pose_iter_584000.caffemodel --save_engine body_25_fp16.engine --run_mode 1 --input ../test.jpg

/testopenpose --prototxt ~/lib/openpose/openpose/models/pose/body_25/pose_deploy.prototxt --caffemodel ~/lib/openpose/openpose/models/pose/body_25/pose_iter_584000.caffemodel --save_engine body_25_fp32.engine --run_mode 0 --input ../test.jpg

fp16和fp32这两个我都试过了。

我的电脑是2070 cuda10.0

还有一个很尴尬的请求，就是我想在我的qt项目用openpose，但是呢，我不是很懂cu文件在qt怎么编译（其实是完全不懂cuda编程），所以希望如果有空的话能不能把openpose弄成so这样.....当然如果没空就算了叭，我还学得动

[ZHEQIUSHUI]

西大普泵，我用qt编译成功了....但是还是没结果唉

[zerollzeng]

只要改动一下CMakeLists.txt就可以编译成动态库,我是为了方便测试才编译成可执行文件的.另外关于没有结果这个问题, 你有更多的信息吗?

[zerollzeng]

我自己在本地测试fp16和fp32都是可以产生正确结果的.大概长这样

[ZHEQIUSHUI]

我把doInfence里的注释解开了，下面是输出。然后现在用的trt是6.0.1.5 cudnn7.6.3

[2019-12-24 11:28:55.425] [info] create plugin factory [2019-12-24 11:28:55.425] [info] yolo3 params: class: 1, netSize: 416 [2019-12-24 11:28:55.425] [info] upsample params: scale: 2 [2019-12-24 11:28:55.426] [info] prototxt: /home/arno/lib/openpose/openpose/models/pose/body_25/pose_deploy.prototxt [2019-12-24 11:28:55.426] [info] caffeModel: /home/arno/lib/openpose/openpose/models/pose/body_25/pose_iter_584000.caffemodel [2019-12-24 11:28:55.426] [info] engineFile: /home/arno/baidunetdiskdownload/tensorrt-zoo/bin/body_25_fp16.engine [2019-12-24 11:28:55.426] [info] outputBlobName: net_output [2019-12-24 11:28:55.426] [info] deserialize engine from /home/arno/baidunetdiskdownload/tensorrt-zoo/bin/body_25_fp16.engine [2019-12-24 11:28:57.362] [info] max batch size of deserialized engine: 1 [2019-12-24 11:28:57.363] [info] create execute context and malloc device memory... [2019-12-24 11:28:57.363] [info] init engine... [2019-12-24 11:28:57.364] [info] malloc device memory nbBingdings: 2 [2019-12-24 11:28:57.364] [info] input: [2019-12-24 11:28:57.364] [info] binding bindIndex: 0, name: image, size in byte: 3072 [2019-12-24 11:28:57.364] [info] binding dims with 3 dimemsion 3 x 16 x 16
[2019-12-24 11:28:57.364] [info] output: [2019-12-24 11:28:57.364] [info] binding bindIndex: 1, name: net_output, size in byte: 1248 [2019-12-24 11:28:57.364] [info] binding dims with 3 dimemsion 78 x 2 x 2
=====>malloc extra memory for openpose... heatmap Dims3 heatmap size: 1 78 2 2 allocate heatmap host and divice memory done resize map size: 1 78 8 8 kernel size: 1 78 8 8 allocate kernel host and device memory done peaks size: 1 25 128 3 allocate peaks host and device memory done =====> malloc extra memory done [2019-12-24 11:28:57.369] [info] net forward takes 4.93421 ms

for (int i =0;i< 10;i++) { std::cout << net_output[i] << " "; }

对应

0.000358858 0.000556114 8.63859e-05 0.000114613 0.00025778 0.000307868 0.000192473 0.000267894 0.00071007 0.00062031

std::cout << std::endl; for(int i=0;i<6080;i++) { std::cout << int(net_output[i]255/25); if((i+1)%80 == 0) { std::cout << std::endl; } }

对应

00000000000000000000000000000000000000000000000000000000000000000000000000000000 000000000000000000001010101000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000000000000000000000000000000000 number of person: 0

[ZHEQIUSHUI]

看起来可能有点累眼睛哈....然后我的图片就是

[zerollzeng]

你要修改prototxt里面的

name: "OpenPose - BODY_25"
input: "image"
input_dim: 1 # This value will be defined at runtime
input_dim: 3
input_dim: 480 # This value will be defined at runtime
input_dim: 640 # This value will be defined at runtime

input dim 为480和640,然后删除掉生成的引擎文件,应该可以出来正确的结果.

详细可以参考readme

[ZHEQIUSHUI]

啊 对不起 我的错 我记得我看了这个的 但是没太在意..太谢谢你啦

[ZHEQIUSHUI]

我 close啦 再次谢谢你！