search

yijingru / BBAVectors-Oriented-Object-Detection

[WACV2021] Oriented Object Detection in Aerial Images with Box Boundary-Aware Vectors
MIT License
462 stars 87 forks source link

Maybe a error about heatmap #106

Closed Lg955 closed 2 years ago

Lg955 commented 2 years ago

In this code, No variable accepts return value: https://github.com/yijingru/BBAVectors-Oriented-Object-Detection/blob/7efd410f5f4ded94aca986aeb0bd1292235d1314/datasets/base.py#L184 So the "hm" is zeros all the time? https://github.com/yijingru/BBAVectors-Oriented-Object-Detection/blob/7efd410f5f4ded94aca986aeb0bd1292235d1314/datasets/base.py#L246

heatmap

yijingru commented 2 years ago

it's in-place, you may visualize using cv2.imshow(normalization(hm)*255)

Lg955 commented 2 years ago

it's in-place, you may visualize using cv2.imshow(normalization(hm)*255)

OK,I will try it after the training, thank U!

Lg955 commented 2 years ago

I check the heatmap in two places(both in datasets/base.py ): 1. hm1 the result is: hm2

2. hm4 the result is: (the bottom left maybe a correct heatmap) hm3

So I want to know whether my method and result are correct?

yijingru commented 2 years ago

As far as I know, draw_umich_gaussian() is fine to create ground truth. But you can make it return to ensure correctness.