app.py

import gradio as gr
import earthview as ev
import utils
import random
import pandas as pd
import os
from itertools import islice

# Configuration
chunk_size = 100  # Size of the chunks to shuffle
label_file = os.path.join(os.path.dirname(__file__), "labels.csv")  # Save CSV in the same directory as the script

# Load the Satellogic dataset (streaming)
dataset = ev.load_dataset("satellogic", streaming=True)
data_iter = iter(dataset)
shuffled_chunk = []  # Initialize an empty list to hold the current chunk
chunk_iter = None  # Initialize the chunk iterator

# Initialize or load labels DataFrame
labels_df = None
if os.path.exists(label_file):
    labels_df = pd.read_csv(label_file)
else:
    labels_df = pd.DataFrame(columns=["image_id", "bounds", "rating", "google_maps_link"])

def get_next_image():
    global data_iter, labels_df, shuffled_chunk, chunk_iter

    while True:
        # If we don't have a current chunk or it's exhausted, get a new one
        if not shuffled_chunk or chunk_iter is None:
            chunk = list(islice(data_iter, chunk_size))
            if not chunk:  # If the dataset is exhausted, reset the iterator
                print("Dataset exhausted, resetting iterator.")
                reset_dataset_iterator() # Use the reset function
                chunk = list(islice(data_iter, chunk_size))
                if not chunk:
                    print("Still no data after reset.")
                    return None, "Dataset exhausted", None, None

            random.shuffle(chunk)
            shuffled_chunk = chunk
            chunk_iter = iter(shuffled_chunk)

        try:
            sample = next(chunk_iter)
            sample = ev.item_to_images("satellogic", sample)
            image = sample["rgb"][0]
            metadata = sample["metadata"]
            bounds = metadata["bounds"]
            google_maps_link = utils.get_google_map_link(sample, "satellogic")
            image_id = str(bounds)

            if labels_df is not None and image_id not in labels_df["image_id"].values:
                return image, image_id, bounds, google_maps_link
            elif labels_df is None: # Handle case where labels_df is not initialized yet
                return image, image_id, bounds, google_maps_link
        except StopIteration:
            # Current chunk is exhausted, reset chunk variables to get a new one in the next iteration
            shuffled_chunk = []
            chunk_iter = None

def rate_image(image_id, bounds, rating):
    global labels_df

    new_row = pd.DataFrame(
        {
            "image_id": [image_id],
            "bounds": [bounds],
            "rating": [rating],
            "google_maps_link": [""], # this isn't necessary to pass to the function since we aren't updating it here.
        }
    )
    labels_df = pd.concat([labels_df, new_row], ignore_index=True)
    labels_df.to_csv(label_file, index=False)

    next_image, next_image_id, next_bounds, next_google_maps_link = get_next_image()
    return next_image, next_image_id, next_bounds, next_google_maps_link

def save_labels_parquet():
    global labels_df
    if labels_df is not None and not labels_df.empty:
        table = pa.Table.from_pandas(labels_df)
        pq.write_table(table, 'labeled_data.parquet')
        return 'labeled_data.parquet'
    else:
        return None

def reset_dataset_iterator():
    global data_iter, shuffled_chunk, chunk_iter
    data_iter = iter(ev.load_dataset("satellogic", streaming=True))
    shuffled_chunk = []
    chunk_iter = None

def load_different_batch():
    print("Loading a different batch of images...")
    reset_dataset_iterator()
    return get_next_image() # Return the first image from the new batch

# Gradio interface
with gr.Blocks() as iface:
    image_out = gr.Image(label="Satellite Image")
    image_id_out = gr.Textbox(label="Image ID", visible=False)
    bounds_out = gr.Textbox(label="Bounds", visible=False)
    google_maps_link_out = gr.Textbox(label="Google Maps Link", visible=True)
    rating_radio = gr.Radio(["Cool", "Not Cool"], label="Rating")
    with gr.Row():
        submit_button = gr.Button("Submit Rating")
        different_batch_button = gr.Button("Load Different Batch") # New button
        download_button = gr.Button("Download Labels (Parquet)")
    download_output = gr.File(label="Download Labeled Data")

    submit_button.click(
        fn=rate_image,
        inputs=[image_id_out, bounds_out, rating_radio],
        outputs=[image_out, image_id_out, bounds_out, google_maps_link_out],
    )

    download_button.click(
        fn=save_labels_parquet,
        inputs=[],
        outputs=[download_output],
    )

    different_batch_button.click(
        fn=load_different_batch,
        inputs=[],
        outputs=[image_out, image_id_out, bounds_out, google_maps_link_out]
    )

     # Load the first image
    initial_image, initial_image_id, initial_bounds, initial_google_maps_link = get_next_image()

    # Set initial values
    if initial_image:
        iface.load(lambda: (initial_image, initial_image_id, initial_bounds, initial_google_maps_link),
                   inputs=None,
                   outputs=[image_out, image_id_out, bounds_out, google_maps_link_out])

iface.launch(share=True)