search

hbin0701 / VT_with_NR_for_FER

[ICASSP 2022] Vision Transformer Equipped with Neural Resizer on Facial Expression Recognition Task
7 stars 1 forks source link

dic images sizes #2

Open Antoniodevy opened 5 months ago

Antoniodevy commented 5 months ago

what are the configs options used for the DIC since i didn't understand what it means with *4 . what is the size before and after for the images ?

hbin0701 commented 5 months ago

Hi, as far as I remember, it should work as long as you make the input images larger than the required input 224 x 224 for Vision Transformers. (i.e. Since Fer+ is 32x32, you should apply x8 to make it 256x256. For RAF-DB, since we used aligned RAF-DB (100x100), you should apply x4 to make it 400x400.)