search

twke18 / Adaptive_Affinity_Fields

Adaptive Affinity Fields for Semantic Segmentation
https://jyhjinghwang.github.io/projects/aaf.html
Other
262 stars 62 forks source link

problem about training the train.py #5

Open zxm115 opened 6 years ago

zxm115 commented 6 years ago

screenshot_2018-09-29 tensorboard_segloss screenshot_2018-09-29 tensorboard_reducedloss screenshot_2018-09-29 tensorboard_l2loss Sorry, I am currently loading the pre-trained model according to the parameter settings in the paper and the sample file named train_pspnet.sh in the bashscripts directory, also using the PASCAL VOC data set, but the same problem still occurs, the loss function can not be reduced, resulting in the prediction result of the verification data set is not correct,which it can't display the content, I would like to ask you how to solve this situation, thank you for your help.

sunkyya commented 6 years ago

@zxm115 I have the same problem with you,have you solved it? I used deeplabV3+，added aaf loss to the original loss,but failed. The original loss can not converge and the result is nonsence.