victory comments

Results 25 comments of


                                            victory

out of memory

Same problem. After 5000 steps, the problem occurs.

I found that by reducing the anchors can somehow alleviate this problem, you can reduce some angels or ratios in R-DFPN_FPN_Tensorflow/libs/configs/cfgs.py ANCHOR_ANGLES = [-90, -75, -60, -45, -30, -15] ANCHOR_RATIOS...

CUDA_ERROR_ILLEGAL_ADDRESS error when many objects in one image

Thanks for your share. When i control the objects per image, it finally works.

The problem for not fixing BN

I have tried the following changes(I only train the network on voc2007 and test on voc2007): 1. Fix BN parameters : The result does not improve a lot, only 1%(The...

The problem for not fixing BN

I have tried the faster-rcnn written by endernewton(tf-faster-rcnn). I have been using this code for my research and project. Training on voc2007, the test result on voc 2007 is 75.7%...

The problem for not fixing BN

Thanks.

convert_data_to_tfrecord.py error!

Make sure that your cfgs.DATASET_NAME in the /lib/config/cfgs.py has corresponding NAME_LABEL_MAP in the /lib/label_name_dict and the key in the NAME_LABEL_MAP dict contains all object class in your own dataset.

Training process occurs nan at the first ten batch.

@ymjiang @bobzhuyb

Training process occurs nan at the first ten batch.

Do not use byteps, the code training process works.