Darknet: Can not get maP & Loss Chart on browser.

[miteshgithub]

I have compiled your repo with GPU=1, CUDNN=1 and OPENCV=1 .
I have amazon ec2 instance.
I have fired this command for training

./darknet detector train ../yolo/darknet.data ../yolo/darknet-yolov3.cfg ../yolo/weights_scales/darknet-yolov3_last.weights -gpus 0,1,2,3 -letter_box -dont_show -mjpeg_port 8090 -map >> ../yolo/train_full_scales.log &

My train_full_scales.log

(next mAP calculation at 4000 iterations)
403: 491.547485, 911.329041 avg loss, 0.000000 rate, 17.527661 seconds, 103168 images
MJPEG-stream sent.
Loaded: 0.000134 seconds

I tried on browser but it does not respond.

[AlexeyAB]

It seems you should open port 8090 in your Amazon EC2 instance: https://postmarkapp.com/support/article/1026-resolving-aws-port-25-throttling

[ambr89]

Hi,
I train my dataset with this code on my own pc:

./darknet detector train data/my.data cfg/my_yolov3-tiny-brav.cfg data/brav/darknet53.conv.74 -map

How can I see the chart? Where I found it?

[AlexeyAB]

@ambr89
It will create chart.png file for each 100 iterations

You must compile Darknet with OpenCV (set OPENCV=1 in the Makefile)

[aluds123]

@AlexeyAB
Hi, I have already train my model on my PC.
But I miss my chart.png, can I recreate it by doing something?

Thanks.

[AlexeyAB]

@aluds123 No.

[Sharon507]

It seems you should open port 8090 in your Amazon EC2 instance: https://postmarkapp.com/support/article/1026-resolving-aws-port-25-throttling

I was having the same issue and this advice fixed it for me. Thank you! It might be worth adding it to the instructions here: https://github.com/AlexeyAB/darknet/issues/1380#issuecomment-412333942.

Sharon

[AlexeyAB]

@Sharon507 I added it.

[Sharon507]

@Sharon507 I added it.

Thank you!

[ZambranoAd]

I have opened an EC2 instance (GPU p3.2x large), but when I try to run the make command on the Makefile for OPENCV=1, it states that the OPENCV package is missing.

I am focused on tinyYOLOv3 training and testing. If I could compile Darknet with OPENCV on the remote GPU, then I would be able to check the loss chart and mAP values.

I would appreciate your help and assistance on this matter

[Sharon507]

Hi @ZambranoAd . Which AMI are you using? I used Deep Learning Base AMI (Ubuntu 16.04). See here for details (including the openCV version installed): https://aws.amazon.com/marketplace/pp/B077GCZ4GR?ref=cns_srchrow.

I had no problems building with GPU=1 CUDNN=1 CUDNN_HALF=1 OPENCV=1.

[ZambranoAd]

Hi @ZambranoAd . Which AMI are you using? I used Deep Learning Base AMI (Ubuntu 16.04). See here for details (including the openCV version installed): https://aws.amazon.com/marketplace/pp/B077GCZ4GR?ref=cns_srchrow.

I had no problems building with GPU=1 CUDNN=1 CUDNN_HALF=1 OPENCV=1.

Thank you very much Sharon507, I was compiling OpenCV through CMake on a Linux Instance but it would have taken so much time. Once again thank you for your quick response