How to set end to end train？

[engineer1109]

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

[auroua]

The lib/model/train_val.py is training the total_loss, and the lib/model/train_val_rfcn_iteration4.py is four steps iteration. You can examine which module is imported in the trainval_net.py file.

[engineer1109]

@auroua from nets.resnet_v1_rfcn_hole_local_global_elements_wise import resnetv1 can‘t found resnet_v1_rfcn_hole_local_global_elements_wise I have done what you said

[auroua]

Because resnet_v1_rfcn_hole_local_global_elements_wise have poor performance, so this file is not included in this repository. You should modify this line as follows: from net.resnet_v1_rfcn_hole import resnetv1

[engineer1109]

2017-12-23 15:53:15.708552: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] @auroua what is it？

[auroua]

I don't know. Please print the full stack trace.

[engineer1109]

iter: 1380 / 250000, total loss: 0.197462

rpn_loss_cls: 0.000246 rpn_loss_box: 0.000517 loss_cls: 0.190126 loss_box: 0.006573 lr: 0.001000 speed: 0.378s / iter 2017-12-23 16:00:28.784699: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:29.064467: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:29.440686: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:29.759453: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:30.196283: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:30.529760: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:30.894242: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:31.220191: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:31.561026: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:31.906444: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:32.259867: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:32.627881: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:32.957452: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:33.284437: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:33.629424: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:33.971635: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:34.317390: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:34.688736: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:35.022211: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24] 2017-12-23 16:00:35.349839: I tensorflow/core/kernels/logging_ops.cc:79] smooth l1 loss outputs[300 24][300 24][300 24][300 24][300 24][300 24]

It spammed in the console again and again. @auroua

[engineer1109]

@auroua It looks like it was about res101_rfcn_ohem.yml When I use the res101_rfcn.yml, it disappeared.

[auroua]

This is caused by this line L266. I will add the details of how to train the network in readme file.