search

dmaniry / deepIQA

NR and FR IQA models based on Deep Convolutional Neural Networks
MIT License
204 stars 75 forks source link

Release training code #1

Open weizhou-geek opened 7 years ago

weizhou-geek commented 7 years ago

Hi,

Your work is very interesting and thanks for sharing the codes! However, I cannot find the training codes. Can you share with me? I would like to ask how the FR and NR models construct, using what kind of framework, like caffe, tensorflow and so on?

dmaniry commented 7 years ago

Currently, I did not have the time to clean up the training code for release. But the models, as defined in fr_model.py and nr_model.py, contain anything necessary for training using the chainer framework and your own data loading code. All details regarding training should be documented in the paper.

AakashKumarNain commented 7 years ago

Hi @dmaniry. Great work!! I am interested in the implementation of NR IQA method, however in your code, you have only defined the forward pass and not any backward pass. Can you provide a snippet of your training code?

stalagmite7 commented 7 years ago

@dmaniry , I've been trying to implement the training using the details provided in the paper, but the model doesn't converge at all. Would be VERY helpful if you could put up the training code so we can have a look at it and use it for our implementation! Thanks, really appreciate it!