Caffe: what's the difference between SSD500 and SSD512?
Hi, @weiliu89
I noticed that you method on PASCAL VOC Challenge 2012 is SSD512，so what's the difference between SSD500 and SSD512? Is there just resized image size or there are something else different?
ShadowLau
All 4 comments
512 is just nice because every time you decrease size by 2. Result wise, it should be similar to 500 if everything else is same.
Actually I just updated the SSD512 (with COCO) result today. It has 82.2 mAP on VOC2012 test with single model on a single image and no post-processing steps. It has the following major change:
- no pool6, every additional layers are convolutional.
- color distortion
- expand the image to get more smaller training examples.
- not clipping the prior box at the boundary.
I will update the arXiv paper later.
weiliu89
on 11 Oct 2016
👍10
excellent！
looking forward to that.
ShadowLau
on 11 Oct 2016
@weiliu89 hello, can you provide the SSD512 python script training on coco ?thank you
yanxp
on 23 Nov 2016
@yanxp I will provide the code later.
weiliu89
on 24 Nov 2016
👍5
Was this page helpful?
0 / 5 - 0 ratings
Related issues
fnzhan
·
5Comments
Jiaqiww
·
5Comments
alenaliu
·
3Comments
lcj1105
·
3Comments
zhang11wu4
·
4Comments
Most helpful comment
512 is just nice because every time you decrease size by 2. Result wise, it should be similar to 500 if everything else is same.
Actually I just updated the SSD512 (with COCO) result today. It has 82.2 mAP on VOC2012 test with single model on a single image and no post-processing steps. It has the following major change:
I will update the arXiv paper later.