Trained until final weights but still no detection

[baptistesa]

Hello,

I try to detect 6 custom objects. For each object, I have 100 pictures, so 600 pictures in my images folder. For the cfg file, I use 6 as the number of class, then 33 for filters parameter.

I used Google Colab for the training, and after ~15H of training and 6000 iterations (which seems to be low), I get the final.weights file. But when I try to use this file to detect my custom objects on trained pictures (retrieved from my images folder), no bounding boxes are drawn. I used Google Colab also for the detection.

I don't know exactly what the issue is, if it's due to a small amount of pictures or something else.

Is there a method to know what's exactly the issue ?

Here is the output of the detection:

Thanks a lot !

[dhiren-hamal]

This problem is related to displaying the image from the docker. The docker container trying to connect to the host display or ubuntu X server. Solution:

1. While creating the container you have to use flag -e DISPLAY=$DISPLAY I.e, nvidia-docker run -d -it -v /home/readib/Experiments/:/Experiments -p 8888:8888 -v /tmp/.X11-unix/:/tmp/.X11-unix -e DI SPLAY=$DISPLAY nvidia/cuda:8.0-cudnn6-devel /bin/bash

2. You need to give permission to xhost xhost +

3. Then try to run code to show the image inside the docker container I hope it will be helpful for someone.