app.py

from datasets import load_dataset
import gradio as gr
import os
import random

wmtis = load_dataset("nlphuji/wmtis")['test']
dataset_size = len(wmtis)

IMAGE = 'image'
IMAGE_DESIGNER = 'image_designer'
DESIGNER_EXPLANATION = 'designer_explanation'
CROWD_CAPTIONS = 'crowd_captions'
CROWD_EXPLANATIONS = 'crowd_explanations'
CROWD_NEGATIVE_CAPTIONS = 'crowd_negative_captions'
CROWD_NEGATIVE_EXPLANATIONS = 'crowd_negative_explanations'
QA = 'question_answering_pairs'
IMAGE_ID = 'image_id'
dataset_cols = [x for x in wmtis.features.keys() if x not in [IMAGE]]
enumerate_cols = [CROWD_CAPTIONS, CROWD_EXPLANATIONS, CROWD_NEGATIVE_CAPTIONS, CROWD_NEGATIVE_EXPLANATIONS]
def func(index):
    example = wmtis[index]
    values = get_instance_values(example)
    return values


def get_instance_values(example):
    values = []
    for k in [IMAGE] + dataset_cols:
        if k in enumerate_cols:
            value = list_to_string(example[k])
        elif k == QA:
            qa_list = [f"Q: {x[0]} A: {x[1]}" for x in example[k]]
            value = list_to_string(qa_list)
        else:
            value = example[k]
        values.append(value)
    return values
def list_to_string(lst):
    return '\n'.join(['{}. {}'.format(i+1, item) for i, item in enumerate(lst)])

demo = gr.Blocks()

with demo:
    gr.Markdown("# Slide to iterate WMTIS")

    with gr.Column():
        slider = gr.Slider(minimum=0, maximum=dataset_size, step=1, label='index')
        with gr.Row():
            index = random.choice(range(0, dataset_size))
            image_input = gr.Image(value=wmtis[index]["image"])
            with gr.Column():
                example = wmtis[index]
                instance_values = get_instance_values(example)
                text_inputs = []
                assert len(dataset_cols) == len(instance_values[1:]) # excluding the image
                for key, value in zip(dataset_cols, instance_values[1:]):
                    label = key.capitalize().replace("_", " ")
                    text_input_k = gr.Textbox(value=value, label=label)
                    text_inputs.append(text_input_k)

    slider.change(func, inputs=[slider], outputs=[image_input] + text_inputs)

demo.launch()