Built a real-time Pipe Detection system with YOLOv8

Excited to share my latest project: Building a Computer Vision Model with YOLOv8 to Count Pipes! 🚀

As a machine learning enthusiast, I recently dove into object detection to solve a real-world problem—accurately counting pipes in industrial settings. Using YOLOv8, I trained a custom model on a dataset sourced from Google Images, and the results were impressive: the model successfully detects and counts pipes with high precision!

The training process was seamless on Google Colab, where I handled data augmentation, model fine-tuning, and evaluation. Once trained, I deployed the model locally using Flask, creating a simple web app for real-time inference. This setup makes it easy to integrate into workflows for inventory management or quality control.

– **Why**: Speed up inspection/estimation workflows with consistent, explainable detections.
– **What I built**:
– Custom-trained YOLOv8 model on a pipe dataset (train/test/valid)
– Robust label cleaning (segmentation → bounding boxes), Drive-safe handling
– Google Colab notebook for GPU training + Drive persistence
– Flask web app: upload from desktop or phone, confidence slider, annotated results, and live pipe count

### Tech stack
– **Model**: Ultralytics YOLOv8 (PyTorch)
– **Training**: Google Colab + Google Drive integration
– **App**: Flask, OpenCV, Pillow, NumPy

### What you can do with it
– Upload a photo from your phone or desktop
– Adjust the confidence threshold
– Get JSON detections, an annotated image, and an automatic pipe count

### Results
– Out-of-the-box YOLO struggled on pipe images; the fine-tuned model delivers far more reliable detections and counts.
– The web demo makes it easy for non-technical users to try the model instantly.

Check out the attached video demo below, showcasing the training in Colab and a quick run of the Flask app. (Note: The video includes some audio glitches with repeated “Heat” mentions—likely from background noise during recording, but the visuals tell the story!)

Related: Built a Safety Equipment Detection using YOLO.

Related: Face detection using google vision.