search

jimeffry / MTCNN-TF

face and pedestrian detection using the MTCNN to train
8 stars 8 forks source link

can't find file about train_Net.py #1

Closed junedgar closed 6 years ago

junedgar commented 6 years ago

hi, thank you for your share.

I couldn't find the file about train_Net.py which you described in prepare_data/gen12_run.sh.

jimeffry commented 6 years ago

first, I am sorry about that. I have a lot work to do, so I don't have enough time to complete the file ReadMe.md. You will find the file train_Net.py in ./Project_root_dir/train_models/. If you have any other questions, please write here, thankyou.

junedgar commented 6 years ago

@Lxyzgf
hi , thank you for your reply! I encountered another problem. When train data put into the P_Net_W, it got the error: ValueError: Negative dimension size caused by subtracting 3 from 2 for 'conv3/Conv2D' (op: 'Conv2D') with input shapes: [256,2,2,64], [3,3,64,128]. The input images' size is 12x12，but there do three times pooling with strade 2 which lead to Negative dimension occurred.

1532409699296

jimeffry commented 6 years ago

@junedgar Hi, the P_Net_W is designed for task of Pedestrian detection. And I find that the MTCNN is not suitable for Pedestrian detection. If you plan to implement the task of Pedestrian detection, I suggest that you do it in other ways, for example Faster-RCNN or FCN. Actually, the input size of P_Net_W maybe 48x48. For the task of Face detection, please use P_Net. Thankyou, for your attention.

junedgar commented 6 years ago

Get it, thank you for reply !