search

zhaoweicai / mscnn

Caffe implementation of our multi-scale object detection framework
404 stars 211 forks source link

Which model do you use to produce the result in the WIDER Face Result page? #103

Closed fengyuentau closed 5 years ago

fengyuentau commented 5 years ago

There are 3 models under examples/widerface, which are mscnn-12x, mscnn-12x-trainval and cascade-mscnn-12s-align. Judging from the commit time, it cannot be the last one which is used to produce the result in the WIDER Face Result page.

Since your paper does not include any details about experiments on WIDER Face, could you tell us which model you use to produce the result in the WIDER Face Result page? Also, by comparing the two files (mscnn_delopy.prototxt) from the two directories, I noticed there are some differences in the detection subnet part. Could you also tell us why they are different?

Thanks.

zhaoweicai commented 5 years ago

mscnn-12s and mscnn-12s-trainval are identical, but with different training dataset, train and trainval respectively. And their mscnn_delopy.prototxt are identical. cascade-mscnn-12s-align is for our work of Cascade R-CNN.

fengyuentau commented 5 years ago

@zhaoweicai Thanks, closing this issue.