import streamlit as st from pathlib import Path import matplotlib.pyplot as plt import numpy as np from sklearn.metrics import confusion_matrix from app.inference import load_models, predict_image, evaluate_dataset from app.config import DATA_ROOT, IMAGE_EXTENSIONS # Paths inside the Space PROJECT_ROOT = Path(__file__).resolve().parent # ------------------------------------------------------------------- # STATIC REPORT TEXTS # ------------------------------------------------------------------- TRAINING_REPORT_TEXT = """ precision recall f1-score support pins_Adriana Lima 1.00 1.00 1.00 32 pins_Alex Lawther 1.00 1.00 1.00 23 pins_Alexandra Daddario 0.97 1.00 0.99 34 pins_Alvaro Morte 1.00 1.00 1.00 30 pins_Amanda Crew 1.00 1.00 1.00 30 pins_Andy Samberg 1.00 1.00 1.00 29 pins_Anne Hathaway 0.97 1.00 0.98 30 pins_Anthony Mackie 1.00 1.00 1.00 30 pins_Avril Lavigne 1.00 0.96 0.98 24 pins_Ben Affleck 1.00 1.00 1.00 30 pins_Bill Gates 1.00 1.00 1.00 30 pins_Bobby Morley 0.97 1.00 0.98 30 pins_Brenton Thwaites 0.97 0.97 0.97 31 pins_Brian J. Smith 1.00 1.00 1.00 30 pins_Brie Larson 1.00 0.92 0.96 25 pins_Chris Evans 1.00 1.00 1.00 25 pins_Chris Hemsworth 1.00 1.00 1.00 24 pins_Chris Pratt 1.00 1.00 1.00 26 pins_Christian Bale 1.00 1.00 1.00 23 pins_Cristiano Ronaldo 1.00 1.00 1.00 30 pins_Danielle Panabaker 0.96 1.00 0.98 27 pins_Dominic Purcell 1.00 1.00 1.00 30 pins_Dwayne Johnson 1.00 1.00 1.00 30 pins_Eliza Taylor 1.00 1.00 1.00 24 pins_Elizabeth Lail 1.00 1.00 1.00 23 pins_Emilia Clarke 1.00 0.97 0.98 31 pins_Emma Stone 1.00 0.97 0.98 30 pins_Emma Watson 0.94 1.00 0.97 32 pins_Gwyneth Paltrow 1.00 1.00 1.00 28 pins_Henry Cavil 1.00 1.00 1.00 29 pins_Hugh Jackman 1.00 0.96 0.98 27 pins_Inbar Lavi 1.00 1.00 1.00 30 pins_Irina Shayk 0.96 1.00 0.98 23 pins_Jake Mcdorman 1.00 1.00 1.00 24 pins_Jason Momoa 1.00 1.00 1.00 28 pins_Jennifer Lawrence 0.96 1.00 0.98 27 pins_Jeremy Renner 1.00 1.00 1.00 25 pins_Jessica Barden 1.00 0.93 0.97 30 pins_Jimmy Fallon 1.00 1.00 1.00 30 pins_Johnny Depp 1.00 1.00 1.00 27 pins_Josh Radnor 1.00 1.00 1.00 30 pins_Katharine Mcphee 1.00 1.00 1.00 27 pins_Katherine Langford 1.00 1.00 1.00 34 pins_Keanu Reeves 1.00 1.00 1.00 24 pins_Krysten Ritter 1.00 1.00 1.00 26 pins_Leonardo DiCaprio 0.94 0.97 0.96 35 pins_Lili Reinhart 0.96 1.00 0.98 22 pins_Lindsey Morgan 0.93 1.00 0.96 25 pins_Lionel Messi 1.00 1.00 1.00 30 pins_Logan Lerman 1.00 0.97 0.98 32 pins_Madelaine Petsch 0.97 1.00 0.98 29 pins_Maisie Williams 1.00 1.00 1.00 29 pins_Maria Pedraza 1.00 1.00 1.00 30 pins_Marie Avgeropoulos 1.00 1.00 1.00 24 pins_Mark Ruffalo 1.00 1.00 1.00 27 pins_Mark Zuckerberg 1.00 1.00 1.00 30 pins_Megan Fox 1.00 1.00 1.00 31 pins_Miley Cyrus 1.00 1.00 1.00 27 pins_Millie Bobby Brown 0.97 1.00 0.98 29 pins_Morena Baccarin 1.00 1.00 1.00 26 pins_Morgan Freeman 1.00 1.00 1.00 30 pins_Nadia Hilker 1.00 1.00 1.00 30 pins_Natalie Dormer 1.00 1.00 1.00 29 pins_Natalie Portman 1.00 1.00 1.00 25 pins_Neil Patrick Harris 1.00 1.00 1.00 30 pins_Pedro Alonso 1.00 1.00 1.00 30 pins_Penn Badgley 1.00 1.00 1.00 26 pins_Rami Malek 1.00 1.00 1.00 24 pins_Rebecca Ferguson 1.00 1.00 1.00 27 pins_Richard Harmon 1.00 0.97 0.98 30 pins_Rihanna 1.00 1.00 1.00 30 pins_Robert De Niro 1.00 1.00 1.00 23 pins_Robert Downey Jr 1.00 1.00 1.00 35 pins_Sarah Wayne Callies 1.00 0.96 0.98 24 pins_Selena Gomez 1.00 0.96 0.98 28 pins_Shakira Isabel Mebarak 1.00 1.00 1.00 23 pins_Sophie Turner 1.00 1.00 1.00 30 pins_Stephen Amell 1.00 1.00 1.00 24 pins_Taylor Swift 1.00 1.00 1.00 30 pins_Tom Cruise 1.00 1.00 1.00 29 pins_Tom Hardy 1.00 1.00 1.00 30 pins_Tom Hiddleston 1.00 1.00 1.00 27 pins_Tom Holland 1.00 0.96 0.98 28 pins_Tuppence Middleton 1.00 1.00 1.00 30 pins_Ursula Corbero 1.00 1.00 1.00 25 pins_Wentworth Miller 1.00 1.00 1.00 27 pins_Zac Efron 1.00 1.00 1.00 29 pins_Zendaya 1.00 0.97 0.98 30 pins_Zoe Saldana 1.00 0.96 0.98 28 pins_alycia dabnem carey 1.00 1.00 1.00 32 pins_amber heard 1.00 1.00 1.00 33 pins_barack obama 1.00 1.00 1.00 30 pins_barbara palvin 1.00 1.00 1.00 29 pins_camila mendes 1.00 1.00 1.00 24 pins_elizabeth olsen 1.00 0.94 0.97 33 pins_ellen page 0.93 1.00 0.97 28 pins_elon musk 1.00 1.00 1.00 30 pins_gal gadot 0.97 0.97 0.97 30 pins_grant gustin 1.00 1.00 1.00 27 pins_jeff bezos 1.00 1.00 1.00 30 pins_kiernen shipka 0.97 1.00 0.98 30 pins_margot robbie 0.97 0.97 0.97 33 pins_melissa fumero 1.00 1.00 1.00 23 pins_scarlett johansson 1.00 0.97 0.98 30 pins_tom ellis 0.97 1.00 0.99 34 accuracy 0.99 2975 macro avg 0.99 0.99 0.99 2975 weighted avg 0.99 0.99 0.99 2975
5-fold CV: mean=0.9917 std=0.0013 Saved centroids to embeddings_cache\\centroids.npy and classes to embeddings_cache\\classes.npy Centroid baseline accuracy: 0.9870771569745344 Suggested cosine threshold for open-set (approx TPR=0.95): 0.4617 """ PREDICTION_REPORT_TEXT = """Loading trained model... ✅ Model loaded. Can recognize 105 classes Found 17534 images in human_face_dataset/pins_face_recognition Processing images... Predicting: 100%|██████████████████████████████████████████████████████████████| 17534/17534 [1:42:42<00:00, 2.85it/s] ================================================================================ 📊 PREDICTION SUMMARY ================================================================================ ✅ Total images processed: 17486 ✅ Correct predictions: 17442 (99.75%) ❌ Wrong predictions: 44 (0.25%) 📊 Overall Accuracy: 0.9975 (99.75%) 📊 Average confidence: 0.8239 ❌ Failed to process: 48 images ================================================================================ ❌ WRONG PREDICTIONS DETAILS (44 total) ================================================================================ Classes with wrong predictions: actual_class wrong_count predicted_as pins_Brie Larson 4 pins_Emma Stone, pins_ellen page, pins_Jennifer Lawrence pins_Jessica Barden 3 pins_Alex Lawther, pins_kiernen shipka, pins_Danielle Panabaker pins_Logan Lerman 3 pins_Sarah Wayne Callies, pins_Leonardo DiCaprio, pins_Eliza Taylor pins_scarlett johansson 2 pins_gal gadot, pins_Taylor Swift pins_Emilia Clarke 2 pins_Marie Avgeropoulos, pins_Irina Shayk pins_Brenton Thwaites 2 pins_Leonardo DiCaprio, pins_Bobby Morley pins_elizabeth olsen 2 pins_ellen page, pins_Millie Bobby Brown pins_Zendaya 2 pins_Lili Reinhart, pins_Selena Gomez pins_Tom Holland 2 pins_Anne Hathaway, pins_Robert Downey Jr pins_Tom Hiddleston 1 pins_Chris Evans pins_Selena Gomez 1 pins_Emma Watson pins_Taylor Swift 1 pins_Emma Stone pins_Zoe Saldana 1 pins_Lindsey Morgan pins_Richard Harmon 1 pins_Leonardo DiCaprio pins_amber heard 1 pins_Brie Larson pins_gal gadot 1 pins_Madelaine Petsch pins_margot robbie 1 pins_Emma Watson pins_Sarah Wayne Callies 1 pins_Lindsey Morgan pins_Avril Lavigne 1 pins_Alexandra Daddario pins_Natalie Portman 1 pins_Millie Bobby Brown pins_Nadia Hilker 1 pins_Marie Avgeropoulos pins_Marie Avgeropoulos 1 pins_Johnny Depp pins_Lindsey Morgan 1 pins_Anne Hathaway pins_Leonardo DiCaprio 1 pins_Brenton Thwaites pins_Katherine Langford 1 pins_Madelaine Petsch pins_Johnny Depp 1 pins_Leonardo DiCaprio pins_Irina Shayk 1 pins_Maria Pedraza pins_Hugh Jackman 1 pins_tom ellis pins_Emma Stone 1 pins_margot robbie pins_Chris Hemsworth 1 pins_Chris Evans pins_Morena Baccarin 1 pins_camila mendes -------------------------------------------------------------------------------- Individual wrong predictions (showing first 20): -------------------------------------------------------------------------------- • amber heard214_323.jpg Actual: pins_amber heard → Predicted: pins_Brie Larson (confidence: 0.275) • Avril Lavigne238_664.jpg Actual: pins_Avril Lavigne → Predicted: pins_Alexandra Daddario (confidence: 0.131) • Brenton Thwaites46_885.jpg Actual: pins_Brenton Thwaites → Predicted: pins_Leonardo DiCaprio (confidence: 0.273) • Brenton Thwaites99_936.jpg Actual: pins_Brenton Thwaites → Predicted: pins_Bobby Morley (confidence: 0.204) • Brie Larson157_994.jpg Actual: pins_Brie Larson → Predicted: pins_Emma Stone (confidence: 0.136) • Brie Larson172_1007.jpg Actual: pins_Brie Larson → Predicted: pins_ellen page (confidence: 0.253) • Brie Larson187_1021.jpg Actual: pins_Brie Larson → Predicted: pins_Jennifer Lawrence (confidence: 0.127) • Brie Larson77_1088.jpg Actual: pins_Brie Larson → Predicted: pins_margot robbie (confidence: 0.330) • Chris Hemsworth1_384.jpg Actual: pins_Chris Hemsworth → Predicted: pins_Chris Evans (confidence: 0.095) • elizabeth olsen164_1173.jpg Actual: pins_elizabeth olsen → Predicted: pins_ellen page (confidence: 0.355) • elizabeth olsen170_1179.jpg Actual: pins_elizabeth olsen → Predicted: pins_Millie Bobby Brown (confidence: 0.272) • Emilia Clarke194_952.jpg Actual: pins_Emilia Clarke → Predicted: pins_Marie Avgeropoulos (confidence: 0.150) • Emilia Clarke48_1021.jpg Actual: pins_Emilia Clarke → Predicted: pins_Irina Shayk (confidence: 0.069) • Emma Stone36_1779.jpg Actual: pins_Emma Stone → Predicted: pins_margot robbie (confidence: 0.282) • gal gadot134_1690.jpg Actual: pins_gal gadot → Predicted: pins_Madelaine Petsch (confidence: 0.234) • Hugh Jackman118_1288.jpg Actual: pins_Hugh Jackman → Predicted: pins_tom ellis (confidence: 0.128) • Irina Shayk236_2335.jpg Actual: pins_Irina Shayk → Predicted: pins_Maria Pedraza (confidence: 0.082) • Jessica Barden211_1449.jpg Actual: pins_Jessica Barden → Predicted: pins_Alex Lawther (confidence: 0.779) • Jessica Barden31_1475.jpg Actual: pins_Jessica Barden → Predicted: pins_kiernen shipka (confidence: 0.098) • Jessica Barden34_1478.jpg Actual: pins_Jessica Barden → Predicted: pins_Danielle Panabaker (confidence: 0.048) ... and 24 more wrong predictions (see CSV for details) ================================================================================ 🎯 CLASSES WITH 100% ACCURACY (74 classes) ================================================================================ class_name total_count pins_Adriana Lima 213 pins_Millie Bobby Brown 191 pins_Rihanna 132 pins_Rebecca Ferguson 178 pins_Rami Malek 160 pins_Penn Badgley 171 pins_Pedro Alonso 125 pins_Neil Patrick Harris 116 pins_Natalie Dormer 196 pins_Morgan Freeman 102 pins_Miley Cyrus 178 pins_Keanu Reeves 158 pins_Megan Fox 208 pins_Mark Zuckerberg 95 pins_Mark Ruffalo 177 pins_Alex Lawther 152 pins_Maisie Williams 193 pins_Madelaine Petsch 192 pins_Lionel Messi 86 pins_Lili Reinhart 150 ... and 54 more classes ================================================================================ ⚠️ CLASSES WITH LOWEST ACCURACY (Bottom 10) ================================================================================ class_name correct_count wrong_count total_count accuracy pins_Taylor Swift 129 1 130 0.992308 pins_elizabeth olsen 219 2 221 0.990950 pins_Brenton Thwaites 207 2 209 0.990431 pins_Emilia Clarke 207 2 209 0.990431 pins_scarlett johansson 199 2 201 0.990050 pins_Tom Holland 187 2 189 0.989418 pins_Logan Lerman 209 3 212 0.985849 pins_Zendaya 135 2 137 0.985401 pins_Jessica Barden 138 3 141 0.978723 pins_Brie Larson 165 4 169 0.976331 ================================================================================ ✅ CORRECT PREDICTIONS SAMPLE (showing 10 of 17442) ================================================================================ ✓ Adriana Lima0_0.jpg: pins_Adriana Lima (confidence: 0.787) ✓ Adriana Lima101_3.jpg: pins_Adriana Lima (confidence: 0.946) ✓ Adriana Lima102_4.jpg: pins_Adriana Lima (confidence: 0.907) ✓ Adriana Lima103_5.jpg: pins_Adriana Lima (confidence: 0.752) ✓ Adriana Lima104_6.jpg: pins_Adriana Lima (confidence: 0.886) ✓ Adriana Lima105_7.jpg: pins_Adriana Lima (confidence: 0.779) ✓ Adriana Lima106_8.jpg: pins_Adriana Lima (confidence: 0.794) ✓ Adriana Lima107_9.jpg: pins_Adriana Lima (confidence: 0.930) ✓ Adriana Lima108_10.jpg: pins_Adriana Lima (confidence: 0.902) ✓ Adriana Lima109_11.jpg: pins_Adriana Lima (confidence: 0.375) ================================================================================ 📁 OUTPUT FILES SAVED: ================================================================================ ✅ predictions_results.csv → All predictions sorted (correct first, then wrong) → Columns: filename, actual, predicted, confidence, status, top3 ✅ predictions_summary.csv → Per-class accuracy summary → Columns: class_name, correct_count, wrong_count, total_count, accuracy ================================================================================ ================================================================================ ❌ FAILED TO PROCESS (48 images) ================================================================================ • Anne Hathaway203_391.jpg: No face detected • Avril Lavigne11_572.jpg: No face detected • Avril Lavigne174_619.jpg: No face detected • Avril Lavigne41_685.jpg: No face detected • barbara palvin158_800.jpg: No face detected • Cristiano Ronaldo209_1326.jpg: No face detected • Cristiano Ronaldo226_1333.jpg: No face detected • Eliza Taylor202_775.jpg: No face detected • Elizabeth Lail102_1055.jpg: No face detected • Elizabeth Lail102_1056.jpg: No face detected • Elizabeth Lail194_1117.jpg: No face detected • Emilia Clarke78_1050.jpg: No face detected • Emma Stone73_1817.jpg: No face detected • Hugh Jackman119_1289.jpg: No face detected • jeff bezos112_2049.jpg: No face detected • jeff bezos12_2052.jpg: No face detected • jeff bezos160_2068.jpg: No face detected • jeff bezos178_2073.jpg: No face detected • Jeremy Renner175_2634.jpg: No face detected • Johnny Depp23_1863.jpg: No face detected ... and 28 more ✅ Failed images list saved to: failed_predictions.csv ================================================================================ ✅ PROCESSING COMPLETE! ================================================================================ """ # ------------------------------------------------------------------- # PAGE CONFIG # ------------------------------------------------------------------- st.set_page_config( page_title="Face Recognition System", layout="wide", ) # ------------------------------------------------------------------- # GLOBAL CSS # ------------------------------------------------------------------- st.markdown( """ """, unsafe_allow_html=True, ) # ------------------------------------------------------------------- # LOAD MODELS ONCE # ------------------------------------------------------------------- @st.cache_resource def init_models(): load_models() return True _ = init_models() # ------------------------------------------------------------------- # NAVBAR # ------------------------------------------------------------------- if "page" not in st.session_state: st.session_state["page"] = "Home" nav_row = st.container() with nav_row: spacer_col, nav_col = st.columns([2, 3]) with nav_col: st.markdown('
', unsafe_allow_html=True) col_home, col_train, col_cm, col_pred, col_about = st.columns( [1.0, 1.9, 1.4, 2.0, 1.1] ) with col_home: if st.button("Home"): st.session_state["page"] = "Home" with col_train: if st.button("Training Report"): st.session_state["page"] = "Training Report" with col_cm: if st.button("C_Matrix"): st.session_state["page"] = "C_Matrix" with col_pred: if st.button("Prediction Report"): st.session_state["page"] = "Prediction Report" with col_about: if st.button("About"): st.session_state["page"] = "About" st.markdown("
", unsafe_allow_html=True) # ------------------------------------------------------------------- # TITLE # ------------------------------------------------------------------- title_row = st.container() with title_row: st.markdown( """

Face Recognition System

""", unsafe_allow_html=True, ) st.markdown("---") page = st.session_state["page"] # ------------------------------------------------------------------- # HOME PAGE # ------------------------------------------------------------------- if page == "Home": st.markdown("### 🔍 Face Recognition Demo") st.write( "Use the controls on the left to select an image from the dataset and run prediction." ) st.sidebar.header("Image Selection") if not DATA_ROOT.exists(): st.sidebar.error( f"Dataset folder not found at:\n`{DATA_ROOT}`\n\n" "Make sure the Hugging Face dataset repo is added as a folder in this Space." ) st.warning("Dataset not ready yet.") selected_image_path = None else: person_folders = sorted([f for f in DATA_ROOT.iterdir() if f.is_dir()]) if not person_folders: st.sidebar.warning("No class folders found inside the dataset directory.") selected_image_path = None else: person_names = [folder.name for folder in person_folders] person_option = st.sidebar.selectbox( "Select Person Folder", ["All"] + person_names ) if person_option == "All": image_files = [ p for p in DATA_ROOT.rglob("*") if p.is_file() and p.suffix.lower() in IMAGE_EXTENSIONS ] else: selected_person_path = DATA_ROOT / person_option image_files = [ p for p in selected_person_path.iterdir() if p.is_file() and p.suffix.lower() in IMAGE_EXTENSIONS ] if not image_files: st.sidebar.warning("No images found in the selected folder.") selected_image_path = None else: display_names = [ img.relative_to(DATA_ROOT).as_posix() for img in image_files ] selected_display = st.sidebar.selectbox("Select Image", display_names) selected_image_path = DATA_ROOT / selected_display if selected_image_path is None: st.info("Dataset not ready or no image selected yet.") else: st.sidebar.image( selected_image_path.as_posix(), caption="Selected Image", use_container_width=True, ) col1, col2 = st.columns([1, 1]) with col1: st.subheader("Chosen Image") st.image(selected_image_path.as_posix(), use_container_width=True) with col2: st.subheader("Prediction Output") if st.button("Predict"): result = predict_image(selected_image_path.as_posix()) if result.get("error"): st.error(result["error"]) else: st.success("Prediction complete") st.write("##### Predicted Label") st.write(f"**{result['predicted_label']}**") st.write("##### Confidence") st.write(f"**{result['confidence']:.4f}**") else: st.info("Click **Predict** to run model inference.") # ------------------------------------------------------------------- # TRAINING REPORT PAGE (TEXT ONLY) # ------------------------------------------------------------------- elif page == "Training Report": st.markdown("### Training Analysis Report") st.write("Classification metrics and evaluation summary from your notebook:") st.code(TRAINING_REPORT_TEXT, language="text") # ------------------------------------------------------------------- # CONFUSION MATRIX PAGE # ------------------------------------------------------------------- elif page == "C_Matrix": st.markdown("### Confusion Matrix (subset evaluation)") st.write( "Programmatically computed confusion matrix on a small subset of the dataset." ) with st.spinner("Computing confusion matrix on a subset of the dataset..."): df, acc = evaluate_dataset(images_per_class=2, max_images=150) if df is None or df.empty: st.info("Could not compute confusion matrix: evaluation subset is empty.") else: labels = sorted(df["true_label"].unique()) cm = confusion_matrix(df["true_label"], df["predicted_label"], labels=labels) fig, ax = plt.subplots(figsize=(8, 8)) im = ax.imshow(cm, interpolation="nearest") ax.set_xticks(np.arange(len(labels))) ax.set_yticks(np.arange(len(labels))) ax.set_xticklabels(labels, rotation=90, fontsize=5) ax.set_yticklabels(labels, fontsize=5) ax.set_xlabel("Predicted label") ax.set_ylabel("True label") ax.set_title("Confusion Matrix (subset)") plt.tight_layout() st.pyplot(fig) if acc is not None: st.caption(f"Subset accuracy on this evaluation run: {acc:.4f}") # ------------------------------------------------------------------- # PREDICTION REPORT PAGE # ------------------------------------------------------------------- elif page == "Prediction Report": st.markdown("### Prediction Report") st.write("Full prediction run analysis and per-class performance:") st.code(PREDICTION_REPORT_TEXT, language="text") # ------------------------------------------------------------------- # ABOUT PAGE # ------------------------------------------------------------------- elif page == "About": st.markdown("### About This Project") readme_path = PROJECT_ROOT / "README.md" if readme_path.exists(): readme_text = readme_path.read_text(encoding="utf-8") st.markdown(readme_text) else: st.write( """ This app demonstrates a deployable face recognition system built with a FaceNet backbone (InceptionResnetV1) and an SVM classifier trained on the `face_recognition_dataset` (Pins celebrities). """ ) st.markdown("---") st.write("Developed by **Mr.Karan**")