How to load side information in Tensorflow?

[gortizji]

Hi,

I would like to load the side information and associate it with the correct sample in Tensorflow. What order do the side_info_cifar10N.csv and side_info_cifar100N.csv follow, the one of the PyTorch or the Tensorflow files?

And if they don't come in the Tensorflow order, should I load them like this?

import numpy as np
import pandas as pd

noise_file = np.load('./data/CIFAR-10_human_ordered.npy', allow_pickle=True)
random_label1 = noise_file.item().get('random_label1')

train_ds, test_ds = tfds.load('cifar10', split=['train','test'], as_supervised=True, batch_size = -1)
train_images, train_labels = tfds.as_numpy(train_ds) 

side_info_df = pd.read_csv('side_info_cifar10N.csv')
worker1_id = df['Worker1-id'].to_numpy()

# Reorder side information with correct order
image_order = np.load('image_order_c10.npy')
worker1_id_ordered = worker1_id[image_order // 10]

# Now, the indexing of all arrays matches correctly
first_example = (train_images[0], train_labels[0], worker1_id_ordered[0])

Thank you very much!

[weijiaheng]

Hi,

Thanks for your concerns. To clarify:
(1) The order of side information matches with the PyTorch version;
(2) For the side information, we did not provide per-sample information. And the provided information is w.r.t. 10 images (a small batch), rather than a single image. The reason is that we wish each worker could contribute not too few on the annotations, i.e., at least annotating 10 images.
(3) Suggested solution for making use of the side information is to load the image_order_c10.npy file and obtain the index mapping between PyTorch and Tensorflow versions. Your solution seems to be reasonable to me!

Best,
Jiaheng

[gortizji]

Thanks a lot Jiaheng, that's what I assumed.

FYI, CIFAR-10/100N is now part of tensorflow_datasets. You might want to link to it in your README.md 😄 .

[weijiaheng]

Thanks for the information, Guille! I will add this to README.md.