import gradio as gr from response_db import ResponseDb from response_db import get_code from create_cache import Game_Cache import numpy as np from PIL import Image import pandas as pd import torch import pickle import uuid import nltk nltk.download('punkt') db = ResponseDb() css = """ .chatbot {display:flex;flex-direction:column} .msg {padding:4px;margin-bottom:4px;border-radius:4px;width:80%} .msg.user {background-color:cornflowerblue;color:white;align-self:self-end} .msg.bot {background-color:lightgray} .na_button {background-color:red;color:red} """ get_window_url_params = """ function(url_params) { console.log(url_params); const params = new URLSearchParams(window.location.search); url_params = Object.fromEntries(params); return url_params; } """ quals = {1001:99, 1002:136, 1003:56, 1004:105} from model.run_question_asking_model import return_modules, return_modules_yn question_model, response_model_simul, response_model_gtruth, caption_model = return_modules() question_model_yn, response_model_simul_yn, response_model_gtruth_yn, caption_model_yn = return_modules_yn() class Game_Session: def __init__(self, taskid, yn, hard_setting): self.yn = yn self.hard_setting = hard_setting global question_model, response_model_simul, response_model_gtruth, caption_model global question_model_yn, response_model_simul_yn, response_model_gtruth_yn, caption_model_yn self.question_model = question_model self.response_model_simul = response_model_simul self.response_model_gtruth = response_model_gtruth self.caption_model = caption_model self.question_model_yn = question_model_yn self.response_model_simul_yn = response_model_simul_yn self.response_model_gtruth_yn = response_model_gtruth_yn self.caption_model_yn = caption_model_yn global image_files, images_np, p_y_x, p_r_qy, p_y_xqr, captions, questions, target_questions self.image_files, self.image_np, self.p_y_x, self.p_r_qy, self.p_y_xqr = None, None, None, None, None self.captions, self.questions, self.target_questions = None, None, None self.history = [] self.game_id = str(uuid.uuid4()) self.set_curr_models() def set_curr_models(self): if self.yn: self.curr_question_model, self.curr_caption_model, self.curr_response_model_simul, self.curr_response_model_gtruth = self.question_model_yn, self.caption_model_yn, self.response_model_simul_yn, self.response_model_gtruth_yn else: self.curr_question_model, self.curr_caption_model, self.curr_response_model_simul, self.curr_response_model_gtruth = self.question_model, self.caption_model, self.response_model_simul, self.response_model_gtruth def get_next_question(self): return self.curr_question_model.select_best_question(self.p_y_x, self.questions, self.images_np, self.captions, self.curr_response_model_simul) def get_model_gtruth_response(self, question): return self.response_model_gtruth.get_response(question, self.images_np[0], self.captions[0], self.target_questions, is_a=self.yn) def ask_a_question(input, taskid, gs): # input = gs.get_model_gtruth_response(gs.history[-1]) if input not in ["n/a", "yes", "no"] and input not in gs.curr_response_model_simul.model.config.label2id: html = "
" for m, msg in enumerate(gs.history): cls = "bot" if m%2 == 0 else "user" html += "
{}
".format(cls, msg) html += "
" return html, gs, gr.Dropdown.update(visible=True, value=''), gr.Button.update(visible=True), gr.Button.update(visible=True), gr.Button.update(visible=True), gr.Number.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=False), gr.Button.update(visible=False), gr.Textbox.update(visible=False), gr.HTML.update(visible=True) gs.history.append(input) gs.p_r_qy = gs.curr_response_model_simul.get_p_r_qy(input, gs.history[-2], gs.images_np, gs.captions) gs.p_y_xqr = gs.p_y_x*gs.p_r_qy gs.p_y_xqr = gs.p_y_xqr/torch.sum(gs.p_y_xqr)if torch.sum(gs.p_y_xqr) != 0 else torch.zeros_like(gs.p_y_xqr) gs.p_y_x = gs.p_y_xqr gs.questions.remove(gs.history[-2]) if taskid not in quals: db.add(gs.game_id, taskid, len(gs.history)//2-1, gs.history[-2], gs.history[-1]) gs.history.append(gs.get_next_question()) top_prob = torch.max(gs.p_y_x).item() top_pred = torch.argmax(gs.p_y_x).item() if top_prob > 0.8 or len(gs.history) > 19: gs.history = gs.history[:-1] if taskid not in quals: db.add(gs.game_id, taskid, len(gs.history)//2, f"Guess: Image {top_pred}", "") # write some HTML html = "
" for m, msg in enumerate(gs.history): cls = "bot" if m%2 == 0 else "user" html += "
{}
".format(cls, msg) html += "
" ### Game finished: if top_prob > 0.8 or len(gs.history) > 19: html += f"

The model identified Image {top_pred+1} as the image. Please select a new task ID to continue.

" finish_html = "

Congratulations on finishing the game! Please copy the Task Finish Code below to MTurk to complete your task. You can now exit this window.

" return html, gs, gr.Dropdown.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=False), gr.HTML.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=False), gr.Button.update(visible=False), gr.Textbox.update(value=get_code(taskid, gs.history, top_pred), visible=True), gr.HTML.update(value=finish_html, visible=True) else: if not gs.yn: return html, gs, gr.Dropdown.update(visible=True, value=''), gr.Button.update(visible=True), gr.Button.update(visible=True), gr.Button.update(visible=True), gr.Number.update(visible=False), gr.HTML.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=False), gr.Button.update(visible=False), gr.Textbox.update(visible=False), gr.HTML.update(visible=False) else: return html, gs, gr.Dropdown.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=False), gr.HTML.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=True), gr.Button.update(visible=True), gr.Textbox.update(visible=False), gr.HTML.update(visible=False) def set_images(taskid): pilot_study = pd.read_csv("pilot-study.csv") taskid_original = taskid if int(taskid) in quals: taskid = quals[int(taskid)] taskid = pilot_study['mscoco-id'].tolist()[int(taskid)] with open(f'cache-soft/{int(taskid)}.p', 'rb') as fp: game_cache = pickle.load(fp) gs = Game_Session(int(taskid), game_cache.yn, game_cache.hard_setting) id1 = f"./mscoco-images/val2014/{game_cache.image_files[0]}" id2 = f"./mscoco-images/val2014/{game_cache.image_files[1]}" id3 = f"./mscoco-images/val2014/{game_cache.image_files[2]}" id4 = f"./mscoco-images/val2014/{game_cache.image_files[3]}" id5 = f"./mscoco-images/val2014/{game_cache.image_files[4]}" id6 = f"./mscoco-images/val2014/{game_cache.image_files[5]}" id7 = f"./mscoco-images/val2014/{game_cache.image_files[6]}" id8 = f"./mscoco-images/val2014/{game_cache.image_files[7]}" id9 = f"./mscoco-images/val2014/{game_cache.image_files[8]}" id10 = f"./mscoco-images/val2014/{game_cache.image_files[9]}" gs.image_files = [id1, id2, id3, id4, id5, id6, id7, id8, id9, id10] gs.image_files = [x[15:] for x in gs.image_files] gs.images_np = [np.asarray(Image.open(f"./mscoco-images/{i}")) for i in gs.image_files] gs.images_np = [np.dstack([i]*3) if len(i.shape)==2 else i for i in gs.images_np] gs.p_y_x = (torch.ones(10)/10).to(gs.curr_question_model.device) gs.captions = gs.curr_caption_model.get_captions(gs.image_files) gs.questions, gs.target_questions = gs.curr_question_model.get_questions(gs.image_files, gs.captions, 0) gs.curr_question_model.reset_question_bank() gs.curr_question_model.question_bank = game_cache.question_dict first_question = gs.curr_question_model.select_best_question(gs.p_y_x, gs.questions, gs.images_np, gs.captions, gs.curr_response_model_simul) first_question_html = f"
{first_question}
" gs.history.append(first_question) # html = f"

Current Task ID: {int(taskid_original)}

" if not gs.yn: return id1, id1, id2, id3, id4, id5, id6, id7, id8, id9, id10, gs, first_question_html, gr.Dropdown.update(visible=True, value=''), gr.Button.update(visible=True), gr.Button.update(visible=True), gr.Button.update(visible=True), gr.Number.update(visible=False), gr.HTML.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False) else: return id1, id1, id2, id3, id4, id5, id6, id7, id8, id9, id10, gs, first_question_html, gr.Dropdown.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=False), gr.Number.update(visible=False), gr.HTML.update(visible=False), gr.Button.update(visible=False), gr.Button.update(visible=True), gr.Button.update(visible=True) def reset_dropdown(): return gr.Dropdown.update(visible=True, value='') with gr.Blocks(title="Image Q&A Guessing Game", css=css) as demo: gr.HTML("

Image Q&A Guessing Game

\

\ Imagine you are playing 20-questions with an AI model.
\ The AI model plays the role of the question asker. You play the role of the responder.
\ There are 10 images. Your image is Image 1. The other images are distraction images.\ The model can see all 10 images and all the questions and answers for the current set of images. It will ask a question based on the available information.
\ The goal of the model is to accurately guess the correct image (i.e. Image 1) in as few turns as possible.
\ Your goal is to help the model guess the image by answering as clearly and accurately as possible.


\ (Note: We are testing multiple game settings. In some instances, the game will be open-ended, while in other instances, the answer choices will be limited to yes/no.)

\ Selecting N/A:
\

\
") with gr.Column(): with gr.Row(): taskid = gr.Number(label="Task ID (Enter a number from 0 to 160)", visible=False) with gr.Row(): start_instruction = gr.HTML("

Press the start button to begin.

") with gr.Row(): start_button = gr.Button("Start") with gr.Column() as img_block: with gr.Row(): img1 = gr.Image(label="Image 1", show_label=True) img2 = gr.Image(label="Image 2", show_label=True) img3 = gr.Image(label="Image 3", show_label=True) img4 = gr.Image(label="Image 4", show_label=True) img5 = gr.Image(label="Image 5", show_label=True) with gr.Row(): img6 = gr.Image(label="Image 6", show_label=True) img7 = gr.Image(label="Image 7", show_label=True) img8 = gr.Image(label="Image 8", show_label=True) img9 = gr.Image(label="Image 9", show_label=True) img10 = gr.Image(label="Image 10", show_label=True) conversation = gr.HTML() game_session_state = gr.State() # answer = gr.Textbox(placeholder="Insert answer here.", label="Answer the given question.", visible=False) full_vocab_dict = response_model_simul_yn.model.config.label2id vocab_list_numbers, vocab_list_letters = [], [] for i in full_vocab_dict: if i=="None" or i is None: continue if i[0] in ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']: vocab_list_numbers.append(i) else: vocab_list_letters.append(i) with gr.Row(): answer = gr.Dropdown(vocab_list_letters+vocab_list_numbers, label="Answer the given question.", \ info="If you cannot find your exact answer, pick the word you feel would be most appropriate. ONLY SELECT N/A IF THE QUESTION DOES NOT APPLY TO THE IMAGE.", visible=False) clear_box = gr.Button("Reset Selection \n(Use this to clear the dropdown selection.)", visible=False) with gr.Row(): vocab_warning = gr.HTML("

The word you typed in is not a valid word in the model vocabulary. Please clear it and select a valid word from the dropdown menu.

", visible=False) null_answer = gr.Textbox("n/a", visible=False) yes_answer = gr.Textbox("yes", visible=False) no_answer = gr.Textbox("no", visible=False) with gr.Column(): with gr.Row(): yes_box = gr.Button("Yes", visible=False) no_box = gr.Button("No", visible=False) with gr.Column(): with gr.Row(): na_box = gr.Button("N/A", visible=False, elem_classes="na_button") submit = gr.Button("Submit", visible=False) with gr.Row(): reward_code = gr.Textbox("", label="Task Finish Code", visible=False) with gr.Column() as img_block0: with gr.Row(): img0 = gr.Image(label="Image 1", show_label=True).style(height=700, width=700) ### Button click events start_button.click(fn=set_images, inputs=taskid, outputs=[img0, img1, img2, img3, img4, img5, img6, img7, img8, img9, img10, game_session_state, conversation, answer, na_box, clear_box, submit, taskid, start_instruction, start_button, yes_box, no_box]) submit.click(fn=ask_a_question, inputs=[answer, taskid, game_session_state], outputs=[conversation, game_session_state, answer, na_box, clear_box, submit, taskid, start_instruction, start_button, yes_box, no_box, reward_code, vocab_warning]) na_box.click(fn=ask_a_question, inputs=[null_answer, taskid, game_session_state], outputs=[conversation, game_session_state, answer, na_box, clear_box, submit, taskid, start_instruction, start_button, yes_box, no_box, reward_code, vocab_warning]) yes_box.click(fn=ask_a_question, inputs=[yes_answer, taskid, game_session_state], outputs=[conversation, game_session_state, answer, na_box, clear_box, submit, taskid, start_instruction, start_button, yes_box, no_box, reward_code, vocab_warning]) no_box.click(fn=ask_a_question, inputs=[no_answer, taskid, game_session_state], outputs=[conversation, game_session_state, answer, na_box, clear_box, submit, taskid, start_instruction, start_button, yes_box, no_box, reward_code, vocab_warning]) clear_box.click(fn=reset_dropdown, inputs=[], outputs=[answer]) url_params = gr.JSON({}, visible=False, label="URL Params") demo.load(fn = lambda url_params : gr.Number.update(value=int(url_params['p'])), inputs=[url_params], outputs=taskid, _js=get_window_url_params) demo.launch()