import streamlit as st import pandas as pd import os import hashlib import sqlite3 # Kept for potential local debugging, though we use CSV for persistence from datetime import datetime from pathlib import Path from huggingface_hub import CommitScheduler from localization_eval import evaluate_submission from PIL import Image # --- CONFIGURATION & SETUP --- st.set_page_config( page_title="AI Benchmark Arena", page_icon="🏆", layout="wide", initial_sidebar_state="expanded" ) # --- HUGGING FACE PERSISTENCE SETUP --- DATA_DIR = Path("data") DATA_DIR.mkdir(exist_ok=True) SUBMISSIONS_CSV = DATA_DIR / "submissions.csv" USERS_CSV = DATA_DIR / "users.csv" # Change 'your-username/your-dataset-name' to your actual repo ID repo_id = "your-username/your-private-dataset" scheduler = CommitScheduler( repo_id=repo_id, repo_type="dataset", folder_path=DATA_DIR, path_in_repo="data", every=5, token=os.getenv("HF_TOKEN") ) def init_db(): """Initializes the CSV files if they do not exist in the data directory.""" if not USERS_CSV.exists(): pd.DataFrame(columns=["username", "password"]).to_csv(USERS_CSV, index=False) if not SUBMISSIONS_CSV.exists(): pd.DataFrame(columns=["username", "bbox_mAP", "bbox_AP50", "segm_mAP", "segm_AP50", "timestamp"]).to_csv(SUBMISSIONS_CSV, index=False) def make_hashes(password): return hashlib.sha256(str.encode(password)).hexdigest() def add_user(username, password): with scheduler.lock: df = pd.read_csv(USERS_CSV) if username in df['username'].values: return False new_user = pd.DataFrame([{"username": username, "password": make_hashes(password)}]) df = pd.concat([df, new_user], ignore_index=True) df.to_csv(USERS_CSV, index=False) return True def login_user(username, password): if not USERS_CSV.exists(): return [] df = pd.read_csv(USERS_CSV) user_match = df[(df['username'] == username) & (df['password'] == make_hashes(password))] return user_match.values.tolist() def save_submission(username, bbox_mAP, bbox_AP50, segm_mAP, segm_AP50): with scheduler.lock: df = pd.read_csv(SUBMISSIONS_CSV) new_row = { "username": username, "bbox_mAP": bbox_mAP, "bbox_AP50": bbox_AP50, "segm_mAP": segm_mAP, "segm_AP50": segm_AP50, "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S") } df = pd.concat([df, pd.DataFrame([new_row])], ignore_index=True) df.to_csv(SUBMISSIONS_CSV, index=False) def get_leaderboard_data(): if not SUBMISSIONS_CSV.exists(): return pd.DataFrame() df = pd.read_csv(SUBMISSIONS_CSV) if df.empty: return df # Logic: Get the highest segm_mAP per user, then the earliest timestamp if tied df['timestamp'] = pd.to_datetime(df['timestamp']) df = df.sort_values(by=['segm_mAP', 'timestamp'], ascending=[False, True]) df_best = df.drop_duplicates(subset='username', keep='first') df_best = df_best.rename(columns={'segm_mAP': 'Best_segm_mAP', 'timestamp': 'last_submission'}) return df_best # --- User Interface --- def ui_login_signup(): st.title("Welcome to Benchmark Arena 🏆") tab1, tab2 = st.tabs(["Login", "Sign Up"]) with tab1: st.subheader("Sign In") username = st.text_input("Username", key="login_user") password = st.text_input("Password", type='password', key="login_pass") if st.button("Login"): if login_user(username, password): st.session_state['logged_in'] = True st.session_state['username'] = username st.rerun() else: st.error("Username or Password incorrect") with tab2: st.subheader("Create New Account") new_user = st.text_input("Username", key="new_user") new_pass = st.text_input("Password", type='password', key="new_pass") if st.button("Sign Up"): if add_user(new_user, new_pass): st.success("Account created! Please navigate to Login.") else: st.warning("Username already exists.") def main_app(): # Sidebar Navigation st.sidebar.title(f"Hi, {st.session_state['username']}!") menu = ["Submit Model", "Leaderboard"] choice = st.sidebar.radio("Navigation", menu) st.sidebar.markdown("---") if st.sidebar.button("Logout"): st.session_state['logged_in'] = False st.session_state['username'] = None st.rerun() with st.expander("ℹī¸ Overview of the AI Benchmark Arena"): st.markdown( """ A natural application of computer vision is to assist blind people, whether that may be to overcome their daily visual challenges or break down their social accessibility barriers. BIV-Priv is proposed to preserve a blind person's visual privacy to ensure they can access visual-related tools safely. VizWiz Challenge 2025 is the 1th edition of the Few-Shot Private Object Localization Challenge on the BIV-Priv dataset. To participate in the challenge, you can find instructions on the Challenge website. Note: All the timings on the EvalAI platform are local to your timezone. """ ) try: overview_image = Image.open("src/overview_image.png").resize((600, 600)) st.image(overview_image, caption="Example of an object localization task") except FileNotFoundError: st.warning("Overview image not found in src/ folder.") with st.expander("📐 How is the Score Calculated?"): st.markdown( """ **Terms and Conditions** The images and annotations in this dataset belong to the VizWiz team and are licensed under a Commons Attribution 4.0 International License. """ ) st.markdown("---") if choice == "Submit Model": st.header("🚀 Submit your Predictions") col1, col2 = st.columns([2, 1]) with col1: uploaded_file = st.file_uploader("Choose a JSON file", type="json") if uploaded_file is not None: save_path = f"./{uploaded_file.name}" with open(save_path, "wb") as f: f.write(uploaded_file.getbuffer()) if st.button("Evaluate"): with st.spinner('Calculating score against Ground Truth...'): # Using your custom evaluation function bbox_mAP, bbox_AP50, segm_mAP, segm_AP50 = evaluate_submission("src/biv_query.json", save_path) if all(v is not None for v in [bbox_mAP, bbox_AP50, segm_mAP, segm_AP50]): st.success(f"Results: bbox_mAP: {bbox_mAP:.2f}, bbox_AP50: {bbox_AP50:.2f}, segm_mAP: {segm_mAP:.2f}, segm_AP50: {segm_AP50:.2f}") save_submission(st.session_state['username'], bbox_mAP, bbox_AP50, segm_mAP, segm_AP50) st.balloons() st.success("Submission Successful!") else: st.error("Evaluation failed. Please check your JSON format.") elif choice == "Leaderboard": st.header("🏆 Leaderboard") st.write("Rankings based on the highest segmentation mAP score achieved.") df_leaderboard = get_leaderboard_data() if not df_leaderboard.empty: df_leaderboard.insert(0, 'Rank', range(1, len(df_leaderboard) + 1)) st.dataframe( df_leaderboard, column_config={ "Rank": st.column_config.Column("Rank", width="small"), "username": "Participant", "Best_segm_mAP": st.column_config.NumberColumn("segm_mAP (Primary)", format="%.4f"), "bbox_mAP": st.column_config.NumberColumn("bbox_mAP", format="%.4f"), "bbox_AP50": st.column_config.NumberColumn("bbox_AP50", format="%.4f"), "segm_AP50": st.column_config.NumberColumn("segm_AP50", format="%.4f"), "last_submission": st.column_config.DatetimeColumn("Last Active Submission", format="D MMM YYYY, h:mm a"), }, use_container_width=True, hide_index=True, ) else: st.info("No submissions yet. Be the first to submit your model!") if __name__ == '__main__': init_db() if 'logged_in' not in st.session_state: st.session_state['logged_in'] = False st.session_state['username'] = None if not st.session_state['logged_in']: ui_login_signup() else: main_app()