Car Identifier (Clean)

AI‑powered desktop app to identify cars from photos using local Ollama vision models, then embed searchable metadata directly into your JPGs (no re‑encoding).

Make • Model • Color • Logos • License Plate • AI Summary → saved to EXIF + IPTC

---

Features

• Local inference with Ollama (privacy by default)
• Vision model support (default: qwen2.5vl:32b-q4_K_M)
• Single image and batch processing (with recursive scanning)
• Overwrite policy for existing metadata: skip, overwrite, ask
• High‑Fidelity input toggle and Enhanced Reasoning mode
• Dark, modern Tkinter UI (optionally styled via ttkbootstrap)
• Fast embedding without re‑encoding pixels:
  • EXIF: UserComment stores a JSON snapshot
  • IPTC: Keywords, Title, Caption/Abstract for Lightroom/Photoshop search
• Windows‑bundled exiftool in this folder for reliable metadata writing

Note: Embedding targets JPG/JPEG files. Non‑JPGs are previewed/analyzed but not embedded.

---

Quick Start

1) Requirements

• Python 3.10+
• Ollama running at http://localhost:11434
  • Install: https://ollama.com
  • Pull a vision model: ollama pull qwen2.5vl:32b-q4_K_M
• Windows: exiftool.exe is included in this folder; macOS/Linux users should install exiftool from https://exiftool.org

2) Install dependencies

cd CarIdentifier_GitHub_Clean
pip install -r requirements.txt
# Optional (improves dark themed widgets):
pip install ttkbootstrap

3) Run

python car_identifier_gui.py

---

Using the App

1. Click “Select Image” (single) or “Select Folder” (batch).
2. Options:
   • ✅ Auto Approve: auto‑save metadata after each inference
   • 💾 Embed in JPG: write EXIF+IPTC directly into JPG/JPEG
   • 🖼️ High Fidelity Input: send original image bytes for best OCR/logo reading
   • 🤖 Enhanced Reasoning: persona + focused crop strategy for harder cases
   • 📁 Recursive Scan: include subfolders during batch processing
   • Existing Metadata: skip / overwrite / ask
3. Click “Process Image” or “Batch Process”.
4. Review the results, then Approve to embed metadata.
5. Bottom pane keeps the “Last Identified Image & Results”.

Tips:

• Mouse wheel to zoom, click‑drag to pan.
• Batch mode shows live progress and current file.

---

What Gets Embedded (JPG/JPEG)

• EXIF UserComment: JSON snapshot of parsed identification
• IPTC Keywords: make/model tokens + helpful tags (e.g., Car Make: BMW, M3, Automotive)
• IPTC Caption/Abstract: human‑readable description
• IPTC ObjectName: concise title (e.g., BMW M3)

Fields extracted include:

• Make, Model, Color
• Logos/emblems/text on the car
• License Plate (if visible)
• AI‑Interpretation Summary (~200‑char description)

By design, this Clean build does not produce JSON/XMP sidecars and only embeds into JPGs. If you need XMP sidecars or embedded XMP, use the v2 app in CarIdentifier_v2.

---

Troubleshooting

• Ollama connection failed
  • Ensure Ollama is running: http://localhost:11434
  • Pull the model: ollama pull qwen2.5vl:32b-q4_K_M
• Model not found
  • The app checks models at startup; follow the on‑screen “pull” instructions.
• exiftool not available (non‑Windows)
  • Install exiftool (brew install exiftool or apt-get install exiftool).
• Tkinter missing (Linux)
  • Install Tk: sudo apt-get install python3-tk.
• Pillow/Ollama import errors
  • Reinstall: pip install -r requirements.txt

---

Notes for Power Users

• Default model: qwen2.5vl:32b-q4_K_M (solid OCR/logo reading). Any Ollama vision‑capable model should work.
• High‑Fidelity Input sends original bytes; disable if bandwidth is constrained.
• Enhanced Reasoning uses persona hints and optional detail crops for harder brand/model disambiguation.
• Embedding avoids pixel re‑encoding by leveraging exiftool where available.

---

Acknowledgements

• ExifTool by Phil Harvey (https://exiftool.org)
• Ollama (https://ollama.com)
• Qwen2.5‑VL model family