issue
usutdzxych opened this issue · comments
What causes the model training not to stop
Hello, you need to specify arguments before (which folder contains data...), type python spatial_train.py -h for more details.
I have set all the parameters. I don't think it's a problem with parameter settings
can you send me these parameter ?
!python3 spatial_train.py -p 16 -b 16
Did you extract the features with extract_features.py ?
yes，I tried 4 videos of myself for training, but no epoch appeared
To train 2BiVQA you need to execute End2End_train.py, spatial_train.py is used only to train the spatial pooling module and it works only with images not videos.
The spatial pooling module is trained on Koniq10k database, you can use it directly with End2End_train.py, I will add further arguments and comments to clarify this point. Thanks.
Thank you for your answer; But I had trained with End2End_train.py，it also happens that no epoch appeared
There is no issues with the train.
yes，I tried 4 videos of myself for training, but no epoch appeared
Here you tried 4 videos, while you specify batch_size as 16
I will close this issue, if you still face the some problem you can contact me at atelili@insa-rennes.fr and we can organize a meeting on zoom or google meet.
After reading your reply, I think it may be the problem of environment and data, because my data is run on Google's Colab. Can you upload the Colab version, including the running data?
You can find a brief demo here: https://drive.google.com/drive/folders/1O4xajRa71K7fkmdjGvLDuNbX8ATdZ_hP?usp=sharing
Thank you for your work. I don't know how long your dataset video is? If possible, can you upload part of the video in Colab and extract features
We used 3 datasets, you can find details in our paper.
I have already uploaded 10 videos features in the drive.
I have no more questions. Thank you for your help. Good luck to you.
I want to reproduce your paper, but I can't find the dataset you synthesized from three dataset in the paper, including videos and MOS.Can you send me your new dataset including videos and MOS?
You can find all details in our paper.
"We also used these three datasets to create a fourth dataset, which is the union of them after MOS calibration using the Iterative Nested Least Squares Algorithm (INSLA)"