Hi,
I am just working on the semantic segmentation module from gluon and I have some questions…I am trying to get an IOU from the segments but wanting to use a bounding box instead of pixel segmentation…
- Is it possible to get a bounding box from each segmentation predictions?
- Is it possible to draw a legend alongside the image returned from get_color_pallete() to show what the corresponding colour is for each class label?
I am using the tutorial demo code at https://gluon-cv.mxnet.io/build/examples_segmentation/demo_fcn.html
Thanks
Hi, eharrihill.
How about using ‘pre-trained Mask RCNN models’?
This might be helpful:
https://gluon-cv.mxnet.io/build/examples_instance/demo_mask_rcnn.html
Hi @eharrihill,
It would be helpful if you could clarify your use-case a bit:
- you want to perform segmentation, then detect the blobs and create a bounding box around them?
- if so I recommend using some blob detection from OpenCV and classical computer vision techniques for that.
- If you want bounding boxes, I would encourage you to use an object detection network from the beginning, because otherwise you’d have a hard time getting bounding box if your pixels predicitions are scattered across your image.