yehao comments

Results 98 comments of


                                            yehao

mAP not the official COCO mAP measurement?

you just compare with [email protected]

A small problem in train.py about conf_thres

you are right!

UserWarning: indexing with dtype torch.uint8 is now deprecated, please use a dtype torch.bool instead.

@paojianghu You should replace `obj_mask=obj_mask.bool() # convert int8 to bool` `noobj_mask=noobj_mask.bool() #convert int8 to bool` in utils/utils.py after line 279

How to test with COCO metrics?

just modify iou_thresh in evaluate() function

Not able to reproduce the mAP results after training

When I use #243 proposed method , mAP increase from 51.4% to 54.9%, but still lower than 55.3%. Another problem is TP calculation in utils/utils.py, please refer to #233

In training, why the loss is decreasing, while the val loss is increasing?

Hi, @oarriaga Can you show your training log? I want to know loss after 120k iterations. Thank you in advance!

Difference on SE from paper

1. In efficientnet, there is NO BN in SE. 2. SE should be added between 3x3 dw and 1x1 pw. 3. it is redundant. You can refer to mobilenet v2.

What is the result of this work on small object detection?

One-stage method is not good at small object detection, you can refer to Feature Pyramid Networks paper to look for the reason.

inference time

Can you post your experiment record?

How to use model to detect objects?

@Eniac-Xie Hi, Can you give me some advice? Thank you in advance!