search

Seanlinx / mtcnn

593 stars 264 forks source link

dataset #46

Open YING1992 opened 7 years ago

YING1992 commented 7 years ago

thanks for share your code. I have few questions for training the model. I try to train P/R/O net with 100000 dataset which include pos, part and neg(2:3:6). However, the results seems bad. I just change the learning rate to 0.00001, other params are keeped. Is my dataset is too small for this problem? If it is possible, can you share your dataset?

Seanlinx commented 7 years ago

@YING1992 How did you get your data set, did you follow the instructions in README to generate training set for Rnet and Onet?

YING1992 commented 7 years ago

thanks for your reply. At first, in my training, i follow the instructions in README. when i generate onet training set, i get nothing with default params. then, i just use method like gen_pnet_data.py to generate rnet and onet training set.

gehaocool commented 6 years ago

Hi, @Seanlinx , I have some questions about preparing the training data from WiderFace.

I followed the data preparation instruction in README

  1. generate data for pnet
  2. use gen_hard_example.py to get data for rnet and onet

After these 2 steps, the amount of rnet training data is: pos: 93576, neg: 4856413, part: 279664, is the pos data enough?

and for onet, pos: 167997, neg:34483, part: 79545, is the negative data and part data enough? I read in another issue that you said 60w false positive is needed for rnet, and 8w for onet. I don't know if I get it right, or I set the wrong params in gen_hard_example.py. Shoud I use the similar method of gen_pnet_data.py to generate some 'easy' samples for rnet and onet?

please give me some instruction, thanks!