mxnet-ssd icon indicating copy to clipboard operation
mxnet-ssd copied to clipboard
verified publisher icon zhreshold

how to futher improve mAP with training from scratch?

Open GarrickLin opened this issue 8 years ago • 2 comments

I have trained a mobilenet based ssd from scratch using trainval set in VOC2007/2012. Now it has an mAP at 0.425539 and seems hard to make large improvement. It might be limited without a pretrained model as I also tried to train reduced vgg16 from scratch. Should I pretrain a good init model with a large number of images of dataset such imageNet as a base knowledgement ? Or use a focalloss to futher improve the performance in defeating imbalance ? I just feel confused and any advice will be appreciated !

GarrickLin avatar Sep 12 '17 05:09 GarrickLin

Recently people have successfully trained a densenet based ssd from scratch, which indicates that with properly designed backpropagate routes, we can skip the pretraining step. The paper is called DSOD: Learning Deeply Supervised Object Detectors from Scratch, FYI.

zhreshold avatar Sep 13 '17 05:09 zhreshold

I have read and implemented DSOD, but the result is not good! Have you tried it? @zhreshold

xhzcyc avatar Oct 11 '17 08:10 xhzcyc