Giters

idealo / image-quality-assessment

Convolutional Neural Networks to predict the aesthetic and technical quality of images.

Home Page:https://idealo.github.io/image-quality-assessment/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to increase accurancy for set from ~ 2k interior images?

Sps113 opened this issue · comments

commented

I tried to train a neural network with a set of about 2 thousand images with user votes according to the aesthetic criterion, about 10 thousand votes.
Json file looks like:
...
{"image_id":"xxxx1","label":[6,1,0,0,1,0,0,1,1,1]},
{"image_id":"xxxx2","label":[0,0,0,1,0,0,0,0,0,1]},
{"image_id":"xxxx3","label":[0,0,0,0,0,0,1,0,0,0]},
{"image_id":"xxxx4","label":[0,0,0,0,0,0,0,0,0,1]},
...
All images from the category of interiors.
After several training attempts with different cleaning data, it was not possible to reduce the prediction error less than 30%.
What I doing wrong?
Please tell me what are the options for increasing the accuracy? Perhaps, need to use other model settings, augmentation, expanding image sets, something else?
I will be grateful for any hint