Just get started caffe, commissioning py - faster - - RCNN today, standing on to run the demo. Py, the following is to run the last few lines of output
I0226 15:10:45. 938380, 6283 net. The CPP: 283] the Network initialization done.
[libprotobuf WARNING Google/protobuf/IO/coded_stream. Cc: 537] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security. Having to happens the limit (or to disable these warnings), see CodedInputStream: : SetTotalBytesLimit () in the Google/protobuf/IO/coded_stream.
h.[libprotobuf WARNING Google/protobuf/IO/coded_stream. Cc: 78] The total number of bytes read was 548317115
I0226 15:10:46. 097867, 6283 net. The CPP: 816] Ignoring the source layer data
I0226 15:10:46. 158936, 6283 net. The CPP: 816] Ignoring the source layer drop6
I0226 15:10:46. 167202, 6283 net. The CPP: 816] Ignoring the source layer drop7
I0226 15:10:46. 167210, 6283 net. The CPP: 816] Ignoring the source layer fc7_drop7_0_split
I0226 15:10:46. 167479, 6283 net. The CPP: 816] Ignoring the source layer loss_cls
I0226 15:10:46. 167484, 6283 net. The CPP: 816] Ignoring the source layer loss_bbox
I0226 15:10:46. 168694, 6283 net. The CPP: 816] Ignoring the source layer silence_rpn_cls_score
I0226 15:10:46. 168699, 6283 net. The CPP: 816] Ignoring the source layer silence_rpn_bbox_pred


The Loaded network/home/XXX/py - faster - - RCNN/data/faster_rcnn_models/VGG16_faster_rcnn_final caffemodel
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
The Demo for data/Demo/000456 JPG
300 object Detection took 0.067 s for proposals
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
The Demo for data/Demo/000542 JPG
259 object Detection took 0.062 s for proposals
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
The Demo for data/Demo/001150 JPG
223 object Detection took 0.050 s for proposals
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
The Demo for data/Demo/001763 JPG
201 object Detection took 0.047 s for proposals
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
The Demo for data/Demo/004545 JPG
172 object Detection took 0.049 s for proposals

But there is no image display, I see a demo. Py finally is PLT. The show () of the code, don't know what went wrong, I tried to try PLT, show () to draw other things can display properly, but I don't know this demo. Py
why not show test result

CodePudding user response:

Know the problem

CodePudding user response:

I also met, this is how to return a responsibility?

CodePudding user response:

I also encountered this problem, how to solve ah, qq2295764512 seek guidance

CodePudding user response:

Hello, I also met the same problem, could you tell me how do you solve?

CodePudding user response:

Originally because of the low recognition rate, if you put the deom. Py the CONF_THRESH change to 0.1, can display images, but the effect will be poor,

CodePudding user response:

I also encountered the same problem yesterday, I found is cuDNN compatibility issues:
On my desktop (GTX1060 graphics) using cuDNN v4 compilation, would be a problem of the building Lord
With cuDNN5.1 was no problem to estimate cuDNN5 also ok (haven't)

Have a problem is not only faster - - RCNN demo, caffe MNIST also calculate out the results

Using notebook (GTX965M) cuDNN v4 and cuDNN5 are normal,

See my blog:
"CuDNN compatibility problems caused by the caffe/mnist, py - faster - - RCNN/demo operation result error" http://blog.csdn.net/10km/article/details/62421445

CodePudding user response:

The building Lord, you are the last question is how to solve? Because I also met this problem

CodePudding user response:

And about 10 km said, I have that in my blog is cuDNN version compatibility

CodePudding user response:

Everybody to solve, I also met this problem, but it seems that the problem, not because CUDNN is the recognition rate is too low

CodePudding user response:

I also encountered this problem, could you tell me how to solve, all of you? Urgent please

CodePudding user response:

Know that the problem is not the problem the reason, also don't say solution, later all don't answer, the building Lord people the most hateful

CodePudding user response:

This is what reason is caused?

CodePudding user response:

Seems not cudnn problem, people know is how to return a responsibility?

CodePudding user response:

reference 5 floor waydong reply:

original because the recognition rate is too low, if you put the deom. Py the CONF_THRESH change to 0.1, you can display images, but the effect will be poor,

hello! CONF_THRESH itself is 0.1, but still don't display picture? thank you

CodePudding user response:

By bloggers and enthusiastic net friend answer, I found the reason of me just because of the low recognition rate is correct detection probability is too low, CONF_THRESH to 0.1 can display images. After just a horse in this category of probability of only 0.110

To summarize, then people can use the following steps of inspection:
1. Change CONF_THRESH to 0.1, ran a try; If not try to 0.01.
2. The above "is in accordance with the said 10 km to modify cudnn," cudnn compatibility problems caused by the caffe/mnist, py - faster - - RCNN/demo operation result error "http://blog.csdn.net/10km/article/details/62421445

CodePudding user response:

Change the CONF_THRESH to 0.01