Using a custom dataset for Detectron2: can't visualize annotations

Hi everyone,
I’m working on a custom dataset using the tutorial use custom dataset.
I’ve have tested (print out) my image annotations and they are all there but I just see the image without annotation overlays using the instruction in the tutorial. Below is the output of an image I’m trying to visualize the annotations. I showed only the first 10 items of the segmentation key.

nb of labels : 3
label 0
bbox : [38, 61, 54, 80]
bbox_mode : 0
category_id : 0
iscrowd : 0
segmentation: [63.5, 62.5, 64.5, 61.5, 66.5, 61.5, 67.5, 62.5, 68.5, 61.5]
label 1
bbox : [64, 85, 62, 61]
bbox_mode : 0
category_id : 1
iscrowd : 0
segmentation: [103.5, 86.5, 104.5, 85.5, 105.5, 85.5, 106.5, 86.5, 107.5, 86.5]
label 2
bbox : [68, 90, 53, 51]
bbox_mode : 0
category_id : 2
iscrowd : 0
segmentation: [92.5, 91.5, 93.5, 90.5, 94.5, 90.5, 95.5, 91.5, 98.5, 91.5]

Some images don’t have my target objects so I set category_id to num_categories (here 3) to indicate that this image is only background (is my use of it correct ??).

As I work in google colab, I’ve tried to modify the source that prompts when accessing the method definition in order to add some print in the overlay_instances. But nothing appears.
Please how can debug this matter.

Thank you so much.

Concerning the local repo modification, I’ve found that the instruction below adapted from the custom dataset tutorial:

!pip install -e "/content/gdrive/My Drive/Colab Notebooks/project_dir"/detectron2_repo

doesn’t make the repo editable at all. No changes in files in the local repo take effect.
So if someone could point me a way to make the repo editable, it’ll help debug my issue.
I’ve searched around and found some other CLI but nothing works:

!cd "/content/gdrive/My Drive/Colab Notebooks/acdc"/detectron2_repo ; python setup.py develop

PS: I’m working on google colab! is that related to my inability to modify the repo ?

Hi, sorry I figured out my dumb tiny mistake: forgot to put “s” at the end the “annotations” key in the data dict.