auroua / tf_rfcn Goto Github PK

I mean you have introduce the training procession but I do not know how to use your model to predict a image.I hope you can give me some suggestions.
And I have another question ,Can I use your code to train my own dataset(I have already made it like VOC2007)

有没有哪个文件是能计算AP、mAP的

请问一下这里边trainval_net和trainval_net_rfcn的区别是什么啊，还有test也有好几个文件，都是干什么用的啊。。。

How to set end to end train？
My train seems four steps
I want to train total loss

How can I train and test on my own dataset?

import _init_paths
You forgot to copy _init_paths.py and import it.
Everything import model.xxxx can't work

when i trained rfcn net with command:
python tools/trainval_net_rfcn.py --cfg=experiments/cfgs/res101_rfcn.yml
I got errors:

......
Loading initial model weights from None
Expected binary or unicode string, got None
Traceback (most recent call last):
File "tools/trainval_net_rfcn.py", line 125, in
max_iters=args.max_iters)
File "/home/aigo/deeplearning/tf_rfcn/lib/model/train_val_rfcn_iteration4.py", line 528, in train_net
sw.train_model(sess, max_iters)
File "/home/aigo/deeplearning/tf_rfcn/lib/model/train_val_rfcn_iteration4.py", line 177, in train_model
variables_to_restore_rpn, variables_to_restore_rfcn = self.net.get_variables_to_restore(variables, var_keep_dic)
File "/home/aigo/deeplearning/tf_rfcn/lib/tf_nets/resnet_v1_rfcn_hole_iterator4.py", line 365, in get_variables_to_restore
if v.name.split(':')[0].replace('rpn_network/', '') in var_keep_dic:
TypeError: argument of type 'NoneType' is not iterable

it may be caused by i have not given a weight file .
what should i do? thanks!

Do your code support for tensorboard？

resnet_v1_rfcn 得到的mAP才0,1,找不到是哪个文件

The accuracy in TensorFlow 1.4.1 is higher than Tensorflow 1.2.1 with this code