hikapok / ron_tensorflow Goto Github PK

Hi,
It confuses me that "Make all anchors on different layers be matched together, to avoid some suboptimal matching results". I cannot figure out what is difference bettween match bbox layer by layer and match bbox together.In fact,we mach every anchor with the gt boxx which has largest overlap with the anchor among all gt bboxes of the input image

hello,May I ask you where to download the vgg_16.ckpt used in your code? Thanks a lot!

when I train ron.py file, this mAP is only 55.97%.how to improve this mAP to 0.74? thanks!

Hello, HiKapok!
I'm so glad to find out your codes to fix up the loss problem which occurs in SSD-Tensorflow. But I wonder how to train the vgg-SSD-Net by your codes. Should I write a new script like ron_net_multi_gpu.py?

HI! HiKapok, thanks for your work, can you share your result on ssd after modification compare to the original paper? Thank you.

Hi there,

Is there a way to view the intermediate objectness prior generated? I believe that it would be great if you could incorporate it.