2025 USA-NA-AIO Round 2, Problem 3, Part 1

AI Olympiads 2025 USA-NA-AIO Round 2
1

Problem 3 (100 points)

In this problem, you are asked to study Contrastive Language-Image Pre-Training (CLIP), a powerful tool in multimodal AI.

# Run code in this cell

"""
DO NOT MAKE ANY CHANGE IN THIS CELL.

HINT: If something is not corrected installed, simply run this cell for few more times.
"""

!pip install datasets transformers

\color{red}{\text{WARNING !!!}}

  • Beyond importing libraries/modules/classes/functions in the following cell, you are NOT allowed to import anything else for the following purposes:

    • As a part of your final solution. For instance, if a problem asks you to build a model without using sklearn but you use it, then you will not earn points.

    • Temporarily import something to assist you to get a solution. For instance, if a problem asks you to manually compute eigenvalues but you temporarily use np.linalg.eig to get an answer and then delete your code, then you violate the rule.

    Rule of thumb: Each part has its particular purpose to intentionally test you something. Do not attempt to find a shortcut to circumvent the rule.

# Run code in this cell

"""
DO NOT MAKE ANY CHANGE IN THIS CELL.
"""

from PIL import Image
import numpy as np
import matplotlib.pyplot as plt
from tqdm import tqdm

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader, Dataset

from transformers import BertTokenizer, BertModel, ViTModel

We will use flickr30k dataset to do image-language matching.

# Run code in this cell

"""

DO NOT MAKE ANY CHANGE IN THIS CELL.

"""

from datasets import load_dataset

dataset_train = load_dataset("USAAIO/2025-Round2-Problem3", split='train')

Part 1 (5 points, coding task)

Do the following tasks to explore the properties of dataset_train:

  1. dataset_train is a list-like object. Print the number of elements in it.

  2. Consider index idx = 2025. Print the type of dataset_train[idx].

  3. Print all keys in dataset_train[idx].

  4. Name the value associated with the key image as image_PIL. Print it.

  5. Convert image_PIL to a NumPy array object, called image_np. Print image_np and its shape.

  6. Display this image by using plt.imshow.

  7. Print the value associated with the key alt_text. Print its type.

2 
### WRITE YOUR SOLUTION HERE ###

print(len(dataset_train))

idx = 2025
print(type(dataset_train[idx]))
print(dataset_train[idx].keys())

image_PIL = dataset_train[idx]['image']
print(image_PIL)

image_np = np.array(image_PIL)
print(image_np)
print(image_np.shape)

plt.imshow(image_np)
plt.show()

print(dataset_train[idx]['alt_text'])
print(type(dataset_train[idx]['alt_text']))

""" END OF THIS PART """