Closed
Description
🚀 Feature
Our utility section just got a revamp! But there are some rough edges.
We added utilities to plot bounding boxes #2556 #2785 and utility to plot semantic segmentation masks #3272 #3330
Pitch
- Add Test to check modifications are not in place. Simply assert if the initial image is the same as final after calling the util.
- For all the utils add docs for Returns. In other docs for boxes.py E.g. nms We specify what it returns.
- Overall code coverage of utils is quite low ~60, bump it up! Most are simple error checks and corner cases.
- Add GitHub gist notebook for all plotting utils. Show a demonstration with torchvision models
- Make the current segmentation utility work for single-channel image. Since mask-rcnn returns masks in single-channel outputs, this utility if extended will serve its job too.
Additional context
The last point should be discussed in a bit detail. Others look straightforward to me.