lakshya-rawat/YOLOV8s-Custom-Logo-detection

YOLOv8s – Custom PDF Logo Detector

Author: lakshya-rawat
Model family: Ultralytics YOLOv8s
Task: Object detection (logo detection on PDF first‑page images)

This repository contains a YOLOv8s model checkpoint trained on a custom logo detection dataset built from the first pages of PDF documents.
It is intended to be used as a building block for applications that need to:

• Detect company logos on PDF first pages
• Use detections as a signal for grouping / sorting PDFs by company or vendor

The main training checkpoint is stored as:

• best.pt – YOLOv8s weights

Note: The surrounding FastAPI application and clustering pipeline are not included here; this repo focuses only on the reusable detection model and a minimal usage script.

---

Model Details

• Architecture: YOLOv8s (Ultralytics)
• Framework: PyTorch via ultralytics Python package
• Input: RGB images (e.g. first page of a PDF rendered to .jpg / .png)
• Output: Bounding boxes, class IDs, confidences for detected logos
• Checkpoint file: best.pt in this repository

Training Plots

Overall training metrics:

Precision / recall curves:

• Precision curve
  
• Recall curve
  
• PR curve
  

Confusion matrices:

• Confusion matrix (absolute)
  
• Confusion matrix (normalized)
  

Intended Use

• Use this model to run logo detection on document images (especially PDF first pages).
• Typical downstream use cases:
  • Clustering / grouping PDFs by detected logo (company)
  • Prioritizing documents with recognized logos

Limitations

• The model was trained on a custom dataset and may not generalize to:
  • Logos outside the training distribution
  • Very different document layouts / scan qualities
• It is not a general object detection model.

---

Installation

Install Python dependencies:

pip install ultralytics torch torchvision

If you plan to use this model from the Hugging Face Hub, also install:

pip install huggingface_hub

---

Quickstart – Local Weights (best.pt)

1. Place best.pt in the same directory as this README.md and main.py.
2. Run inference on an image:

python main.py --weights best.pt --source path/to/first_page_image.jpg

This will:

• Load the YOLOv8s model from best.pt
• Run detection on the provided image
• Display the image with logo bounding boxes overlaid

You can also use the model directly from Python:

from ultralytics import YOLO

model = YOLO("best.pt")  # path to local checkpoint
results = model("path/to/first_page_image.jpg")
results[0].show()  # display image with boxes

---

Using from Hugging Face Hub

Once the model is pushed to the Hub under your account (for example):

• Repo ID: lakshya-rawat/yolov8s-pdf-logo-detector

You can load it directly:

from ultralytics import YOLO

model = YOLO("lakshya-rawat/yolov8s-pdf-logo-detector")
results = model("path/to/first_page_image.jpg")

for r in results:
    r.show()

If you prefer to download the weights first:

huggingface-cli download lakshya-rawat/yolov8s-pdf-logo-detector best.pt
python main.py --weights best.pt --source path/to/first_page_image.jpg

---

Example: Integrating into Your Own App

You can reuse the same pattern as in main.py to plug the model into your own application:

from ultralytics import YOLO

model = YOLO("best.pt")  # or "lakshya-rawat/yolov8s-pdf-logo-detector"

def detect_logos(image_path: str):
    results = model(image_path)
    detections = []
    for r in results:
        for box in r.boxes:
            detections.append({
                "xyxy": box.xyxy[0].tolist(),
                "confidence": float(box.conf[0]),
                "class_id": int(box.cls[0]),
            })
    return detections

You are free to extend this with OCR, clustering, or PDF‑specific logic in your own codebase.

---

Files in This Repo

• best.pt – YOLOv8s checkpoint trained on a custom PDF logo dataset.
• main.py – Minimal example script to run inference via CLI.
• README.md – This model card and usage documentation.