repair to v1.2

app.py

@@ -2,119 +2,155 @@ import gradio as gr
 import cv2
 import whisper
 import torch
-from fer import FER
+import numpy as np
+from PIL import Image
 from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
-# --- 1. LOAD MODELS ---
-print("Sedang memuat model... Mohon tunggu.")
+# --- 1. SETUP & LOAD MODELS ---
+print("Sedang memuat model... Mohon tunggu sebentar.")
 
-# A. Model Otak: SmolLM (Gunakan versi Instruct agar bisa diajak chat/analisis)
-# Kita gunakan versi 135M atau 360M agar ringan di CPU Hugging Face Space gratis
-model_id = "HuggingFaceTB/SmolLM-135M-Instruct" 
+# A. Model Otak: SmolLM (Agent)
+# Menggunakan versi Instruct agar bisa diajak diskusi
+model_id = "HuggingFaceTB/SmolLM-135M-Instruct"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 smol_lm = AutoModelForCausalLM.from_pretrained(model_id)
 
-# B. Model Telinga: Whisper untuk Transkripsi
-whisper_model = whisper.load_model("tiny") # Gunakan 'tiny' atau 'base' agar cepat
+# B. Model Telinga: Whisper (Audio to Text)
+whisper_model = whisper.load_model("tiny")
 
-# C. Model Mata: FER untuk Emosi Wajah
-face_detector = FER(mtcnn=True) # MTCNN lebih akurat
+# C. Model Mata: Vision Transformer untuk Emosi
+# Kita ganti FER dengan model native Hugging Face agar tidak error
+emotion_classifier = pipeline("image-classification", model="dima806/facial_emotions_image_detection")
 
-# --- 2. FUNGSI PEMROSESAN ---
+# D. Setup Deteksi Wajah (OpenCV Basic)
+# Menggunakan Haar Cascade bawaan cv2 untuk menemukan lokasi wajah
+face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
 
-def analyze_emotion(video_path):
-    """
-    Fungsi utama yang memproses video user.
-    """
-    if not video_path:
-        return "Mohon upload video terlebih dahulu."
+# --- 2. FUNGSI LOGIKA ---
 
-    # --- LANGKAH 1: Analisis Audio (Transkripsi) ---
-    # Whisper otomatis ekstrak audio dari file video
-    audio_result = whisper_model.transcribe(video_path)
-    transcribed_text = audio_result["text"]
-    
-    # --- LANGKAH 2: Analisis Visual (Mimik Muka) ---
-    # Kita ambil beberapa frame dari video untuk dicek emosinya
+def get_dominant_emotion(video_path):
     cap = cv2.VideoCapture(video_path)
     emotions_list = []
-    
     frame_count = 0
+    
+    # Ambil sampel setiap 30 frame (sekitar 1 detik sekali)
     while cap.isOpened():
         ret, frame = cap.read()
         if not ret:
             break
         
-        # Cek setiap 30 frame (agar tidak terlalu berat)
         if frame_count % 30 == 0:
-            # FER mendeteksi emosi dominan di frame ini
-            top_emotion, score = face_detector.top_emotion(frame)
-            if top_emotion:
-                emotions_list.append(top_emotion)
+            # 1. Convert ke Grayscale untuk deteksi wajah
+            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+            faces = face_cascade.detectMultiScale(gray, 1.1, 4)
+            
+            for (x, y, w, h) in faces:
+                # 2. Crop bagian wajah saja
+                face_roi = frame[y:y+h, x:x+w]
+                
+                # 3. Convert ke format PIL Image untuk Hugging Face Pipeline
+                rgb_face = cv2.cvtColor(face_roi, cv2.COLOR_BGR2RGB)
+                pil_image = Image.fromarray(rgb_face)
+                
+                # 4. Prediksi Emosi
+                try:
+                    results = emotion_classifier(pil_image)
+                    # results format: [{'label': 'happy', 'score': 0.9}, ...]
+                    top_emotion = results[0]['label']
+                    emotions_list.append(top_emotion)
+                except Exception as e:
+                    print(f"Error detecting frame: {e}")
+                    continue
+                
+                # Kita hanya ambil 1 wajah pertama yang ketemu per frame
+                break 
+                
         frame_count += 1
     cap.release()
+    
+    if not emotions_list:
+        return "Tidak ada wajah terdeteksi"
+    
+    # Cari modus (emosi yang paling sering muncul)
+    return max(set(emotions_list), key=emotions_list.count)
 
-    # Hitung emosi yang paling sering muncul (Modus)
-    if emotions_list:
-        dominant_facial_emotion = max(set(emotions_list), key=emotions_list.count)
-    else:
-        dominant_facial_emotion = "Netral/Tidak Terdeteksi"
+def analyze_agent(video_path):
+    if not video_path:
+        return "Error", "Mohon upload video.", "N/A"
 
-    # --- LANGKAH 3: Analisis Agen (SmolLM3) ---
-    # Kita buat prompt agar SmolLM bertindak sebagai psikolog/analis
+    print(f"Processing video: {video_path}")
+
+    # 1. Transkripsi Audio (Telinga)
+    try:
+        audio_result = whisper_model.transcribe(video_path)
+        transcribed_text = audio_result["text"]
+    except Exception as e:
+        transcribed_text = f"Gagal transkripsi audio: {str(e)}"
+
+    # 2. Deteksi Emosi Visual (Mata)
+    detected_emotion = get_dominant_emotion(video_path)
     
-    system_prompt = "You are an expert AI emotional analyst. Analyze the user's state based on their facial expression and spoken words."
+    # 3. Analisis SmolLM (Otak)
+    system_prompt = "You are an expert AI psychological analyst. Analyze the user's emotion based on facial expression and text."
     
     user_input = f"""
-    DATA INPUT:
-    1. Transcribed Text: "{transcribed_text}"
-    2. Facial Expression Detected: {dominant_facial_emotion}
+    DATA DARI USER:
+    - Teks Ucapan: "{transcribed_text}"
+    - Ekspresi Wajah Dominan: {detected_emotion}
     
-    TUGAS:
-    Jelaskan emosi apa yang dirasakan orang ini? Apakah kata-katanya (teks) cocok dengan ekspresi wajahnya? Berikan kesimpulan singkat dalam Bahasa Indonesia.
+    INSTRUKSI:
+    Analisis apakah ada kesesuaian antara ucapan dan ekspresi wajahnya. 
+    Jika wajah 'sad' tapi teks semangat, mungkin dia menyembunyikan sesuatu.
+    Berikan kesimpulan singkat dalam Bahasa Indonesia.
     """
     
-    # Format prompt sesuai template chat SmolLM
     messages = [
         {"role": "system", "content": system_prompt},
         {"role": "user", "content": user_input},
     ]
     
-    input_text = tokenizer.apply_chat_template(messages, tokenize=False)
-    inputs = tokenizer.encode(input_text, return_tensors="pt")
+    # Format chat template
+    input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt", tokenize=True)
     
-    # Generate jawaban
-    outputs = smol_lm.generate(inputs, max_new_tokens=200, temperature=0.7, top_p=0.9)
-    analysis_result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    # Generate response
+    outputs = smol_lm.generate(input_ids, max_new_tokens=250, temperature=0.7)
+    decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
     
-    # Bersihkan output (hapus prompt asli dari hasil)
-    final_response = analysis_result.split("assistant")[-1].strip()
+    # Parsing output agar rapi (mengambil bagian assistant saja)
+    if "assistant" in decoded:
+        final_response = decoded.split("assistant")[-1].strip()
+    else:
+        # Fallback jika format berbeda
+        final_response = decoded
 
-    return final_response, transcribed_text, dominant_facial_emotion
+    return final_response, transcribed_text, detected_emotion
 
-# --- 3. MEMBUAT UI DENGAN GRADIO ---
+# --- 3. USER INTERFACE ---
 
-with gr.Blocks() as demo:
-    gr.Markdown("# 🧠 SmolLM3 Emotion Agent")
-    gr.Markdown("Upload video pendek (berbicara ke kamera), AI akan mendeteksi: **Mimik Wajah + Teks Ucapan**.")
-    
-    with gr.Row():
-        video_input = gr.Video(label="Upload Video atau Rekam via Webcam")
+css = """
+#col-container {max-width: 700px; margin-left: auto; margin-right: auto;}
+"""
+
+with gr.Blocks(css=css) as demo:
+    with gr.Column(elem_id="col-container"):
+        gr.Markdown("## 🤖 SmolLM3 Multimodal Agent (Video Emotion)")
+        gr.Markdown("Upload video Anda berbicara. AI akan melihat ekspresi wajah dan mendengar ucapan Anda.")
         
-    submit_btn = gr.Button("Analisis Emosi")
-    
-    with gr.Row():
-        output_analysis = gr.Textbox(label="Analisis SmolLM3 (Agent)", lines=5)
-    
-    with gr.Row():
-        output_text = gr.Textbox(label="Teks Terdeteksi")
-        output_face = gr.Textbox(label="Emosi Wajah Dominan")
+        video_input = gr.Video(sources=["upload", "webcam"])
+        submit_btn = gr.Button("Analisis Emosi", variant="primary")
+        
+        gr.Markdown("### Hasil Analisis Agent")
+        output_agent = gr.Textbox(label="Pendapat SmolLM3", lines=4)
+        
+        with gr.Row():
+            output_text = gr.Textbox(label="Transkrip Suara")
+            output_face = gr.Textbox(label="Deteksi Wajah")
 
     submit_btn.click(
-        fn=analyze_emotion, 
-        inputs=video_input, 
-        outputs=[output_analysis, output_text, output_face]
+        fn=analyze_agent,
+        inputs=[video_input],
+        outputs=[output_agent, output_text, output_face]
     )
 
-# Jalankan aplikasi
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -3,8 +3,9 @@ torch
 torchaudio
 gradio
 opencv-python-headless
-fer
 openai-whisper
 numpy
 scipy
-accelerate
+accelerate
+pillow
+timm