fenling / pytorch-ssd

reproduced the proposed results of SSD300 and SSD512

pytorch-ssd

by Qijie Zhao, Feng Ni. (VDIG，PKU)

We won't update this repo any more, welcome to read&use our recent work: (M2Det)[https://github.com/qijiezhao/M2Det]

Reproduced the proposed results.

Besides, our proposed upgraded model will be opened soon!(Note that we are not developed from pure SSD)

• VOC2007

model	mAP
ssd300	77.27%
ssd512	79.89%
Ours300-vgg	80.5%
Ours512-vgg	82.1%
Ours300-resnet101	81.7%
Ours512-resnet101	82.7%

• KITTI

model&Input	mAP
ssd300,TBA	TBA
ssd512,TBA	TBA
Ours300	80.2%
Ours512	82.6%
Ours800	86.7%(==>up to 87.9%)
Ours800-multi-scale	89.83%(==>up to 90.08%)

• MS COCO

model&Input	mAP(0.5:0.95)
Ours300-vgg	30.1%(TBA)
Ours300-resnet101	32.1%
Ours300-vgg-multiscale	36.7%
Ours512	34.8%(TBA)
Ours512-vgg-multiscale	39.0%

Still being under fixing.