arnabdhar/Swin-V2-base-Food
Image Classification
•
Updated
•
22
image
imagewidth (px) 53
224
| label
class label 121
classes |
---|---|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
|
0apple_pie
|
This dataset is the downscaled version of the Food121 dataset. All images are downscaled to a maximum of 224*224.
This dataset is the combination of the Food101, Indian Food Classification and The-massive-Indian-Food-Dataset datasets. This Dataset aims to be a viable dataset for Image Classification of Foods with an added Indian context. This dataset has 121 classes with each class having 800 images in the train split and 200 images in the test split.