search

anson0910 / CNN_face_detection

Implementation based on the paper Li et al., “A Convolutional Neural Network Cascade for Face Detection, ” 2015 CVPR
253 stars 148 forks source link

About the face size in create_face_12c.sh #29

Closed MaxLingwei closed 7 years ago

MaxLingwei commented 7 years ago

Dear anson, I've read you code in create_face_12c.sh. I found that you resize the image to 60 * 60. I don't understand why you do so. And by the way why do you set the RESIZE_HEITHG and RESIZE_WIDTH both to be 15? Hope for your answering. Thank you very much!

anson0910 commented 7 years ago

Hi, the images created from that .sh file are actually 15 * 15, the comments were copied from create_face_48c.sh, sorry.. The face sizes in the lmdb are sizes of 15, such that one image can create 10 crops (top-left, top-right, bottom-left, middle, and mirrors of previous 5 crops) of sizes 12 x 12 during training.

MaxLingwei commented 7 years ago

So you mean that the final image used to training is also 12 * 12?

anson0910 commented 7 years ago

The images created are of size 15 * 15, but caffe will crop blocks of sizes 12 x 12.

MaxLingwei commented 7 years ago

And in which step will caffe crop the image?

anson0910 commented 7 years ago

When you perform training through executing ./build/tools/caffe train --solver=models/face_12c/solver.prototxt

It's specified inside face_12c/train_val.prototxt

MaxLingwei commented 7 years ago

OK, I see. It's very nice for you to answer my question. Thany you very much!