search

atelili / 2BiVQA

2BiVQA is a no-reference deep learning based video quality assessment metric.
MIT License
31 stars 6 forks source link

how to change the input size in pretrained model? #3

Closed Xiaocong6 closed 2 years ago

Xiaocong6 commented 2 years ago

hi~ I would like to do some experiments with your code. I want to change the input size of (30, 25, 2048), but in the pre-trained model 'res-bi-sp_koniq.h5', this input seems to be fixed, how can I get my own pre-trained model with arbitrary size?

atelili commented 2 years ago

Hello, Thank you for your interest in our work. You can change the number of frame N: Input size ( N, 25, 2048). However, you can't change the number of patches (25) yet, but I will add soon a script to do that. Kind regards,

atelili commented 2 years ago

Hello, you can find a script named spatial_train.py. However, it is recommended to train the spatial pooling module on an IQA dataset.