Roman Solovyev

That would be nice if you share your progress in process. I can help with some parts. If you have account at ODS.ai (https://ods.ai/join-community) you can find me there: zfturbo.

Please provide small example.

3D WBF only supports rectangular parallelepipeds as input, given as 2 opposite corners (x1, y1, z1) and (x2, y2, z2). If you can convert your 3D boxes in this format...

Actually algorithm will work without normalization (in this case you need to remove checks and sligtly remade IoU calculation). But anyway it will require rectangular parallelepipeds as input.

Weights needed to say which model predictions are more important. By default all weights equal to 1.

You can use as width max width for all boxes. The same for height.

I didn't have chance to check it. All my models had NMS block at the end. I still have plans to make some related experiments.

Just to be sure about performance on single model, I made an experiment. I took RetinaNet based on ResNet152 backbone which was trained on Open Images dataset. Then I cut...

If any coordinate goes below 0 or higher than 1 it will be trimmed to 0 or 1 respectively. It's in the code here: https://github.com/ZFTurbo/Weighted-Boxes-Fusion/blob/master/ensemble_boxes/ensemble_boxes_wbf.py#L56

> how to normalize? the box coordinate is (x1, y1, x2, y2) which represents the left-top and right-bottom corner, and the image's height is h, image's width is w. the...