Digits failed to draw box after training process

AI & Data Science Deep Learning (Training & Inference) DIGITS (Locked)
1

Hello!
I follow the guide in DIGITS/README.md at master · NVIDIA/DIGITS · GitHub, and finished the training of object detection. However, there is no boxes displayed when I test any picture. Besides, I also cannot find map and accuracy.

The task is finished in PC, Ubuntu16.04

Can you give me some advice? Thanks!

2

This seems due to recently changed flask render_template_string function. If you are using github DIGITS, can you try to change https://github.com/NVIDIA/DIGITS/blob/8a1bbb764f7777d1b646e7c4123cac904861a363/digits/extensions/view/boundingBox/view_template.html#L7 to

$scope.bboxes = {{ bboxes | safe }};

and see if it helps.

If you are using DIGITS container from nvcr.io, the fix will be included in the next release.

3

I used digits from ncvr.io, still faces the same issue…but inference in Jetson displays the BB

4

If you can get into your running container, can you try to do this?

  1. Find the container id with
docker ps

, which is the first column of the results. Let’s call it $DOCKER_ID

  1. Do
docker exec -it $DOCKER_ID /bin/bash
  1. Change the line in /opt/digits/digits/extensions/view/boundingBox/view_template.html to the suggestion in my previous reply.

Sorry for the trouble.

5

I have pulled the latest DIGITS from NGC with ‘docker pull nvcr.io/nvidia/digits:19.12-caffe-py3’ and the inference visualisation with bounding boxes is still not being drawn. Raw data format does show the bounding boxes correctly in text format. I investigated the fix suggested above and it’s already implemented in the container, but bboxes not working.

$scope.bboxes = {{ bboxes | safe }};

Setup is CUDA 10.1 P100 Ubuntu 18.04