Roman Solovyev

You can make inference on mmdetection models and store boxes as lists in memory. Then apply WBF function. WBF is only effective for ensembles or TTA. So it's not replacement...

We can add it, since it's common usage case for NMS. But I already see the error in code. ```final_masks = np.concatenate(final_masks)``` It will fail. And please add comment about...

The easiest solution is to remove predictions from ensemble you don't need to use (e. g. weight = 0).

I meant you can remove second set of boxes and set weights to weights=[1], then WBF will be applied to first set of boxes only. It's the equiualent to weights=[1,...

In our experiments WBF for single model gave worse results than NMS. WBF is good for ensemble of TTA or different models.

As I remeber this was resolved here: https://github.com/ZFTurbo/Weighted-Boxes-Fusion/issues/38 In all experiments it was lower than 1.0. Please try to create counter example.

WBF doesn't support rotated boxes yet. It will be good if you provide some small example.

I used it in Yolov5. What you need to do is to read all txt files with predictions created by Yolo and gather it in single CSV file or just...

You need to convert boxes in x1, y1, x2, y2 format, then normalize (using image width and height). ``` xn1 = x1 / image_width yn1 = y1 / image_height xn2...

You can use this method as is. It has simple code. Or please add more details to your question.