Upload 2 files

app.py

@@ -1,350 +1,44 @@
-# app.py
-import os
-import io
-import uuid
-from datetime import datetime
-from typing import Tuple, Optional
-
 import gradio as gr
-import pandas as pd
+import torch
+import torchvision
+from torchvision import transforms
 from PIL import Image
-from reportlab.lib.pagesizes import A4
-from reportlab.pdfgen import canvas
-from transformers import pipeline
-
-# -------------------------
-# Configuration / Filenames
-# -------------------------
-CSV_FILE = "job_cards.csv"
-EXCEL_FILE = "job_cards.xlsx"
-PDF_FOLDER = "pdf_reports"
-os.makedirs(PDF_FOLDER, exist_ok=True)
-
-# -------------------------
-# Load AI models (pipelines)
-# -------------------------
-# NOTE: these models are reasonably capable but can be large.
-# On Hugging Face Spaces choose a GPU runtime for best performance.
-
-print("Loading models... this may take a while on first run.")
-
-# 1) Text classification (zero-shot)
-text_classifier = pipeline(
-    task="zero-shot-classification",
-    model="facebook/bart-large-mnli"
-)
-
-# 2) Image object detection (DETR)
-image_detector = pipeline(
-    task="object-detection",
-    model="facebook/detr-resnet-50"
-)
-
-# 3) Speech-to-text (Whisper)
-asr = pipeline(
-    task="automatic-speech-recognition",
-    model="openai/whisper-large-v2",
-    chunk_length_s=30
-)
-
-print("Models loaded.")
-
-# -------------------------
-# Knowledge bases (starter)
-# -------------------------
-CATEGORIES = [
-    "Engine issue", "Brake issue", "AC problem", "Steering issue",
-    "Electrical issue", "Regular service", "Body damage", "Tire issue"
-]
-
-REPAIR_JOBS = {
-    "Engine issue": ["Engine diagnosis", "Oil change", "Spark plug check"],
-    "Brake issue": ["Brake pad replacement", "Brake fluid top-up", "Disc inspection"],
-    "AC problem": ["AC gas refill", "Compressor check", "Cabin filter replacement"],
-    "Steering issue": ["Power steering fluid check", "Alignment test"],
-    "Electrical issue": ["Battery test", "Alternator check", "Fuse replacement"],
-    "Regular service": ["Oil change", "Filter replacement", "General inspection"],
-    "Body damage": ["Dent repair", "Paint touch-up", "Bumper replacement"],
-    "Tire issue": ["Tire rotation", "Tire replacement", "Wheel balancing"]
-}
-
-COST_RANGES = {
-    "Engine issue": "₹3000–₹12,000",
-    "Brake issue": "₹1500–₹7000",
-    "AC problem": "₹2500–₹10,000",
-    "Steering issue": "₹1500–₹8000",
-    "Electrical issue": "₹1000–₹5000",
-    "Regular service": "₹1500–₹6000",
-    "Body damage": "₹2000–₹30,000",
-    "Tire issue": "₹800–₹8,000"
-}
-
-# Simple parts mapping (starter)
-PARTS_DB = {
-    "Brake pad replacement": ["Brake pads (set)", "Brake cleaner", "Brake fluid"],
-    "AC gas refill": ["R134a refrigerant", "O-rings", "Compressor oil"],
-    "Oil change": ["Engine oil (4L)", "Oil filter"],
-    "Battery test": ["12V battery", "Battery terminal cleaner"]
-}
-
-# -------------------------
-# Utility functions
-# -------------------------
-def ensure_csv_exists():
-    if not os.path.exists(CSV_FILE):
-        df = pd.DataFrame(columns=[
-            "ID", "Date", "Customer", "Vehicle", "Complaint",
-            "Category", "Suggested_Jobs", "Estimated_Cost",
-            "Detected_Damages", "Parts_Recommendations"
-        ])
-        df.to_csv(CSV_FILE, index=False)
-
-def classify_complaint(complaint: str) -> Tuple[str, str, str]:
-    """Return human text, top category, comma-joined suggested jobs, and cost"""
-    if not complaint or not complaint.strip():
-        return ("❌ No complaint text provided.", "", "", "")
-    res = text_classifier(complaint, candidate_labels=CATEGORIES)
-    top_category = res["labels"][0]
-    suggested_jobs = REPAIR_JOBS.get(top_category, ["Workshop inspection required"])
-    cost = COST_RANGES.get(top_category, "Varies")
-    jobs_text = ", ".join(suggested_jobs)
-    human_text = f"🔍 Category: {top_category}\n📋 Jobs: {jobs_text}\n💰 Estimated cost: {cost}"
-    return human_text, top_category, jobs_text, cost
-
-def detect_damage_from_image(image: Image.Image) -> Tuple[str, str]:
-    """Return human text & comma-joined detections"""
-    if image is None:
-        return ("❌ No image uploaded.", "")
-    pil = image.convert("RGB")
-    results = image_detector(pil, threshold=0.3)
-    # results is a list of dicts with label and score
-    if not results:
-        return ("✅ No obvious damage detected.", "")
-    detections = [f"{r['label']} ({r['score']:.2f})" for r in results]
-    return ("🚗 Detected: " + ", ".join(detections), ", ".join(detections))
-
-def speech_to_text(audio) -> str:
-    """Accept a wav/ogg file-like and return cleaned text"""
-    if audio is None:
-        return ""
-    # the pipeline accepts file path or array; Gradio supplies a filepath
-    try:
-        result = asr(audio)
-        text = result.get("text", "")
-        # simple cleanup
-        return text.strip()
-    except Exception as e:
-        return f"⚠️ ASR error: {e}"
-
-def recommend_parts(jobs_list_str: str, damages_str: str) -> str:
-    """Return a simple list of recommended parts. Placeholder for vector/DB search."""
-    parts = set()
-    if jobs_list_str:
-        # heuristic: check if any known job matches
-        for job_key in PARTS_DB:
-            if job_key.lower() in jobs_list_str.lower():
-                for p in PARTS_DB[job_key]:
-                    parts.add(p)
-    # also check damages keywords
-    if damages_str:
-        if "tire" in damages_str.lower():
-            parts.add("Tire(s)")
-        if "dent" in damages_str.lower():
-            parts.add("Body filler / Paint")
-    if not parts:
-        return "No specific parts suggested. Inspect to confirm."
-    return ", ".join(sorted(parts))
-
-# -------------------------
-# Persistence / Job Card CRUD
-# -------------------------
-ensure_csv_exists()
-
-def save_job_card(
-    customer: str, vehicle: str, complaint: str,
-    category: str, suggested_jobs: str, estimated_cost: str,
-    detected_damages: str, parts: str
-) -> str:
-    if not (customer and vehicle and complaint):
-        return "⚠️ Provide Customer, Vehicle number, and Complaint before saving."
-    df = pd.read_csv(CSV_FILE)
-    new_id = str(uuid.uuid4())[:8]
-    entry = {
-        "ID": new_id,
-        "Date": datetime.now().strftime("%Y-%m-%d %H:%M"),
-        "Customer": customer,
-        "Vehicle": vehicle,
-        "Complaint": complaint,
-        "Category": category,
-        "Suggested_Jobs": suggested_jobs,
-        "Estimated_Cost": estimated_cost,
-        "Detected_Damages": detected_damages,
-        "Parts_Recommendations": parts
-    }
-    df = pd.concat([df, pd.DataFrame([entry])], ignore_index=True)
-    df.to_csv(CSV_FILE, index=False)
-    return f"✅ Saved job card (ID: {new_id})"
-
-def view_all_cards() -> pd.DataFrame:
-    df = pd.read_csv(CSV_FILE)
-    return df
-
-def search_cards(customer_query: str, vehicle_query: str) -> pd.DataFrame:
-    df = pd.read_csv(CSV_FILE)
-    if customer_query:
-        df = df[df["Customer"].str.contains(customer_query, case=False, na=False)]
-    if vehicle_query:
-        df = df[df["Vehicle"].str.contains(vehicle_query, case=False, na=False)]
-    if df.empty:
-        return pd.DataFrame([{"Message": "No matching records found"}])
-    return df
 
-def export_excel() -> Optional[str]:
-    if not os.path.exists(CSV_FILE):
-        return None
-    df = pd.read_csv(CSV_FILE)
-    df.to_excel(EXCEL_FILE, index=False, engine="openpyxl")
-    return EXCEL_FILE
+# Load MobileNetV2 pretrained on ImageNet
+model = torchvision.models.mobilenet_v2(pretrained=True)
+model.eval()
+
+# Transform for input images
+transform = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor(),
+    transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                         std=[0.229, 0.224, 0.225]),
+])
+
+# Load ImageNet class labels
+imagenet_labels = []
+import urllib.request
+url = "https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt"
+imagenet_labels = urllib.request.urlopen(url).read().decode("utf-8").split("\n")
+
+def classify_image(image):
+    img = transform(image).unsqueeze(0)
+    with torch.no_grad():
+        outputs = model(img)
+        probs = torch.nn.functional.softmax(outputs[0], dim=0)
+    top5 = torch.topk(probs, 5)
+    results = {imagenet_labels[idx]: float(probs[idx]) for idx in top5.indices}
+    return results
 
-def generate_jobcard_pdf(job_id: str) -> Optional[str]:
-    df = pd.read_csv(CSV_FILE)
-    row = df[df["ID"] == job_id]
-    if row.empty:
-        return None
-    row = row.iloc[0]
-    filename = os.path.join(PDF_FOLDER, f"jobcard_{job_id}.pdf")
-    buffer = io.BytesIO()
-    c = canvas.Canvas(buffer, pagesize=A4)
-    w, h = A4
-    margin = 50
-    y = h - margin
-
-    c.setFont("Helvetica-Bold", 16)
-    c.drawString(margin, y, f"Job Card — ID: {job_id}")
-    y -= 30
-    c.setFont("Helvetica", 11)
-    lines = [
-        f"Date: {row['Date']}",
-        f"Customer: {row['Customer']}",
-        f"Vehicle: {row['Vehicle']}",
-        f"Complaint: {row['Complaint']}",
-        f"Category: {row['Category']}",
-        f"Suggested Jobs: {row['Suggested_Jobs']}",
-        f"Estimated Cost: {row['Estimated_Cost']}",
-        f"Detected Damages: {row['Detected_Damages']}",
-        f"Parts Recommendations: {row['Parts_Recommendations']}"
-    ]
-    for line in lines:
-        c.drawString(margin, y, line)
-        y -= 18
-        if y < 80:
-            c.showPage()
-            y = h - margin
-            c.setFont("Helvetica", 11)
-    c.save()
-
-    with open(filename, "wb") as f:
-        f.write(buffer.getvalue())
-    return filename
-
-# -------------------------
 # Gradio UI
-# -------------------------
-with gr.Blocks() as demo:
-    gr.Markdown("# 🚘 Full AI Service Advisor — Multimodal (All features)")
-
-    with gr.Tab("Create Job Card"):
-        with gr.Row():
-            with gr.Column(scale=2):
-                customer = gr.Textbox(label="Customer Name")
-                vehicle = gr.Textbox(label="Vehicle Number")
-                complaint = gr.Textbox(label="Complaint (free text)", lines=3, placeholder="e.g. My steering is stiff when turning left...")
-                analyze_btn = gr.Button("Analyze Complaint (NLP)")
-                nlp_out = gr.Textbox(label="Complaint Analysis (NLP output)", lines=4)
-                category_hidden = gr.Textbox(visible=False)
-                jobs_hidden = gr.Textbox(visible=False)
-                cost_hidden = gr.Textbox(visible=False)
-
-                gr.Markdown("### Voice input")
-                audio_input = gr.Audio(label="Record Complaint (or upload audio)", source="microphone", type="filepath")
-                transcribe_btn = gr.Button("Transcribe Audio → Text")
-                transcription_out = gr.Textbox(label="Transcribed Text", lines=3)
-
-            with gr.Column(scale=2):
-                image_input = gr.Image(type="pil", label="Upload car image (damage/photo)")
-                detect_btn = gr.Button("Detect Damage (Image)")
-                detect_out = gr.Textbox(label="Damage Detection (image)", lines=3)
-                det_hidden = gr.Textbox(visible=False)
-
-                parts_out = gr.Textbox(label="Parts Recommendation", lines=2)
-                save_btn = gr.Button("💾 Save Job Card")
-                save_status = gr.Textbox(label="Save status")
-                gen_pdf_btn = gr.Button("📄 Generate PDF for Last Saved")
-                last_pdf_link = gr.File(label="Download PDF")
-
-    with gr.Tab("View / Search / Export"):
-        view_btn = gr.Button("📋 View All Job Cards")
-        df_view = gr.Dataframe()
-
-        gr.Markdown("### Search job cards")
-        cust_search = gr.Textbox(label="Search by Customer name (partial)")
-        veh_search = gr.Textbox(label="Search by Vehicle number (partial)")
-        search_btn = gr.Button("Search")
-        search_out = gr.Dataframe()
-
-        export_btn = gr.Button("Export all as Excel")
-        excel_file = gr.File(label="Download Excel")
-
-    # -----------------------
-    # Callbacks / wiring
-    # -----------------------
-    def on_analyze(complaint_text):
-        human, top_cat, jobs_text, cost_text = classify_complaint(complaint_text)
-        return human, top_cat, jobs_text, cost_text
-
-    analyze_btn.click(on_analyze, inputs=complaint, outputs=[nlp_out, category_hidden, jobs_hidden, cost_hidden])
-
-    def on_transcribe(audio_path):
-        txt = speech_to_text(audio_path)
-        return txt
-
-    transcribe_btn.click(on_transcribe, inputs=audio_input, outputs=transcription_out)
-
-    def on_detect(image_pil):
-        human, detections = detect_damage_from_image(image_pil)
-        return human, detections
-
-    detect_btn.click(on_detect, inputs=image_input, outputs=[detect_out, det_hidden])
-
-    def on_recommend_parts(jobs_text, det_text):
-        parts = recommend_parts(jobs_text, det_text)
-        return parts
-
-    # when either jobs_hidden or det_hidden change, update parts_out (we'll wire after)
-    jobs_hidden.change(on_recommend_parts, inputs=[jobs_hidden, det_hidden], outputs=parts_out)
-    det_hidden.change(on_recommend_parts, inputs=[jobs_hidden, det_hidden], outputs=parts_out)
-
-    def on_save(customer_name, vehicle_no, complaint_text, category_text, jobs_text, cost_text, det_text, parts_text):
-        return save_job_card(customer_name, vehicle_no, complaint_text, category_text, jobs_text, cost_text, det_text, parts_text)
-
-    save_btn.click(on_save, inputs=[customer, vehicle, complaint, category_hidden, jobs_hidden, cost_hidden, det_hidden, parts_out], outputs=[save_status])
-
-    # Generate PDF for the last saved job (we look up last row)
-    def on_generate_pdf():
-        df = pd.read_csv(CSV_FILE)
-        if df.empty:
-            return None
-        last_id = df.iloc[-1]["ID"]
-        path = generate_jobcard_pdf(last_id)
-        if path:
-            return path
-        return None
-
-    gen_pdf_btn.click(on_generate_pdf, outputs=last_pdf_link)
-
-    # View / search / export
-    view_btn.click(lambda: view_all_cards(), outputs=df_view)
-    search_btn.click(lambda a,b: search_cards(a,b), inputs=[cust_search, veh_search], outputs=search_out)
-    export_btn.click(lambda: export_excel(), outputs=excel_file)
+demo = gr.Interface(
+    fn=classify_image,
+    inputs=gr.Image(type="pil"),
+    outputs=gr.Label(num_top_classes=5),
+    title="🚀 Lightweight CPU Image Classifier",
+    description="MobileNetV2-based image classifier optimized for CPU (fast cold starts)."
+)
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -1,8 +1,4 @@
-gradio>=3.0
-transformers
 torch
-timm
-pandas
-openpyxl
-reportlab
+torchvision
+gradio
 Pillow