ious=0

[lu16keqi]

hi! First of all, thank you for publishing this impressive work - thank you very much for your contribution. I ran the training program and found that the ious indicators were all 0, and the gt_box was a 7-element number while the pred_box was a 6-element number. Could it be that there is a problem with the given model file?
I hope you can reply as soon as possible. Thank you very much.

[iris0329]

Hi, thank you for your interest to our work.

The GT bbox has rotation, but our current baseline only outputs 6-dimensional bboxes without rotation. Even so, our method still outperforms the comparison methods.

I’ve been swamped lately, but I’ll verify the code and reply to you soon.

[liuzengyun]

@lu16keqi I've run into the same issue—whether testing the checkpoint provided by the authors or training and testing it myself, the iou is 0. Have you resolved this problem?

[liuzengyun]

Hello @lu16keqi and @iris0329 , I found and fixed this issue where assigning np.float32 values to torch.tensor caused all IOUs to be zeros. pls refer to #5

[iris0329]

hi @liuzengyun, it’s great to hear that! Thanks for your feedback!