Yet-another-efficientdet-pytorch: why threshold 0.05

[ldz666666]

Thanks for your repository
I wonder why in coco_eval,py line 55, the confidence threshold was set to 0.05 ,which seems too low for me and may bring many extra boxes

[alexriedel1]

Yeah you should have as many boxes as possible to effectively find all the true positives and false positives. You could even use 0.00 which might give a slighlty higher mAP increase inference time by a lot