ML / Done
Shale Shaker Cutting Monitoring Web
Overview
Built as my undergraduate thesis (Tugas Akhir): a full migration of the desktop shale-shaker monitoring system to the web. Where v1 used classical OpenCV without deep learning, this version runs deep learning (MobileViT + BiSeNet v2) behind a web tool where users upload a video, mark a 4-point ROI, tune the model, and run live detection with trend visualization.
Challenge
Turning an offline detection model into a usable web tool: video upload, 4-point ROI selection, tunable inference settings, and live trend charts in one session flow.
Result
Delivered a live internal tool (v0.2) running a MobileViT model over ONNX with FastAPI, WebSocket updates, and SQLite session history.
Key Statistics
v0.2 (web migration)
Version
MobileViT
Model
ONNX
Runtime
FastAPI
Backend
WebSocket
Updates
Technologies
Computer Vision
-
Python -
OpenCV - MobileViT
- BiSeNet v2
-
ONNX
Backend
-
FastAPI - WebSocket
-
SQLite
Deployment
-
Docker
Gallery
Detection Dashboard
Live detection view with ROI overlay and coverage trend chart.
Live Detection Session
Real session with ROI overlay, coverage trend, and last coverage value.
