src/label_interface.py

"""Interface for labeling concepts in images.
"""

from typing import Optional

import random

import gradio as gr

from src import global_variables
from src.constants import CONCEPTS, ASSETS_FOLDER, DATASET_NAME


def filter_sample(sample, concepts, username, sample_type):
    has_concepts = all([sample["concepts"].get(c, False) for c in concepts])
    if not has_concepts:
        return False
    if "votes" in sample and username in sample["votes"]:
        is_labelled = all([c in sample["votes"][username] for c in CONCEPTS])
    else:
        is_labelled = False
    if sample_type == "labelled":
        return is_labelled
    elif sample_type == "unlabelled":
        return not is_labelled
    else:
        raise ValueError(f"Invalid sample type: {sample_type}")


def get_next_image(
    split: str,
    concepts: list,
    sample_type: str,
    filtered_indices: dict,
    selected_concepts: list,
    selected_sample_type: str,
    profile: gr.OAuthProfile
):
    username = profile.username
    if concepts != selected_concepts or sample_type != selected_sample_type:
        for key, values in global_variables.all_metadata.items():
            filtered_indices[key] = [i for i in range(len(values)) if filter_sample(values[i], concepts, username, sample_type)]
        selected_concepts = concepts
        selected_sample_type = sample_type
    try:
        sample_idx = random.choice(filtered_indices[split])
        sample = global_variables.all_metadata[split][sample_idx]
        image_path = f"{ASSETS_FOLDER}/{DATASET_NAME}/data/{split}/{sample['file_name']}"
        try:
            username_votes = sample["votes"][username]
            voted_concepts = [c for c in CONCEPTS if username_votes.get(c, False)]
            unseen_concepts = [c for c in CONCEPTS if c not in username_votes]
        except KeyError:
            voted_concepts = []
            unseen_concepts = []
        tie_concepts = [c for c in sample["concepts"] if sample["concepts"][c] is None]

        return (
            image_path, 
            voted_concepts, 
            f"{split}:{sample_idx}", 
            sample["class"], 
            sample["concepts"],
            unseen_concepts,
            tie_concepts,
            filtered_indices, 
            selected_concepts,
            selected_sample_type,
        )
    except IndexError:
        gr.Warning("No image found for the selected filter.")
        return None, None, None, None, None, None, None, filtered_indices, selected_concepts, selected_sample_type

def submit_label(
    voted_concepts: list,
    current_image: Optional[str],
    split, 
    concepts, 
    sample_type,
    filtered_indices, 
    selected_concepts,
    selected_sample_type,
    profile: gr.OAuthProfile
):
    username = profile.username
    if current_image is None:
        gr.Warning("No image selected.")
        return None, None, None, None, None, None, None, filtered_indices, selected_concepts, selected_sample_type
    current_split, idx = current_image.split(":")
    idx = int(idx)
    global_variables.get_metadata(current_split)
    if "votes" not in global_variables.all_metadata[current_split][idx]:
        global_variables.all_metadata[current_split][idx]["votes"] = {}
    global_variables.all_metadata[current_split][idx]["votes"][username] = {c: c in voted_concepts for c in CONCEPTS}
    vote_sum = {c: 0 for c in CONCEPTS}
    new_concepts = {}
    for c in CONCEPTS:
        for vote in global_variables.all_metadata[current_split][idx]["votes"].values():
            if c not in vote:
                continue
            vote_sum[c] += 2 * vote[c] - 1
        new_concepts[c] = vote_sum[c] > 0 if vote_sum[c] != 0 else None
    global_variables.all_metadata[current_split][idx]["concepts"] = new_concepts
    global_variables.save_metadata(current_split)
    gr.Info("Submit success")
    return get_next_image(
        split, 
        concepts,
        sample_type,
        filtered_indices,
        selected_concepts,
        selected_sample_type,
        profile
    )
    

with gr.Blocks() as interface:
    with gr.Row():
        with gr.Column():
            with gr.Group():
                gr.Markdown(
                    "## # Image Selection",
                )
                with gr.Row():
                    split = gr.Radio(
                        label="Split",
                        choices=["train", "validation", "test"],
                        value="train",
                    )
                    sample_type = gr.Radio(
                        label="Sample Type",
                        choices=["labelled", "unlabelled"],
                        value="unlabelled",
                    )
                concepts = gr.Dropdown(
                    label="Concepts",
                    multiselect=True,
                    choices=CONCEPTS,
                )
            with gr.Group():
                voted_concepts = gr.CheckboxGroup(
                    label="Voted Concepts",
                    choices=CONCEPTS,
                )
                unseen_concepts = gr.CheckboxGroup(
                    label="Previously Unseen Concepts",
                    choices=CONCEPTS,
                )
                tie_concepts = gr.CheckboxGroup(
                    label="Tie Concepts",
                    choices=CONCEPTS,
                )

            with gr.Row():
                next_button = gr.Button(
                    value="Next",
                )
                gr.LoginButton()
                submit_button = gr.Button(
                    value="Submit",
                )
            with gr.Group():
                gr.Markdown(
                    "##  # Image Info",
                )
                im_class = gr.Textbox(
                    label="Class",
                )
                im_concepts = gr.JSON(
                    label="Concepts",
                )
        with gr.Column():
            image = gr.Image(
                label="Image",
            )
    current_image = gr.State(None)
    filtered_indices = gr.State({
        split: list(range(len(global_variables.all_metadata[split])))
        for split in global_variables.all_metadata
    })
    selected_concepts = gr.State([])
    selected_sample_type = gr.State(None)
    common_output = [
        image, 
        voted_concepts, 
        current_image, 
        im_class,
        im_concepts,
        unseen_concepts,
        tie_concepts,
        filtered_indices, 
        selected_concepts,
        selected_sample_type,
    ]
    next_button.click(
        get_next_image, 
        inputs=[split, concepts, sample_type, filtered_indices, selected_concepts, selected_sample_type],
        outputs=common_output
    )
    submit_button.click(
        submit_label,
        inputs=[voted_concepts, current_image, split, concepts, sample_type, filtered_indices, selected_concepts, selected_sample_type],
        outputs=common_output
    )