[Proposal] “DarkNet Detect” w/Yolo3

“DarkNet Detect” w/Yolo3 Project Proposal

PROJECT INTRO/SUMMARY

Custom Object detection in single or multiple images themed “DarkNet Detect”
For this project I’ve chosen DarkNet, an open sourced neural network framework that will use pre-trained Yolov3 model weights to detect objects within an image. The image will then be divided into regions while predicting bounding boxes and probabilities for each region based on the COCO Dataset, either 320 or 416 images will be used in the dataset for faster inference times

TECHNICAL CONCEPTS
DarkNet Yolov3, COCO model training Fine Tuning, Dockerize, NGINX, Ainize

GOALS
To detect objects in images via the DarkNet training framework using Yolo3 detection on a COCO Data set with pre-configured weights, developing a docker container and uploading to Docker Hub with a Vue usable container, then Ainize

TIMELINE

• Week 1: Preparing the dataset framework and weight configs, image sizing
• Week 2: Fine tune pre-trained/custom model and determine thresholds
• Week 3: Develop and Dockerize the model app with Vue and NGINX for web use
• Week 4: Upload and serve the Detection app on Docker hub, use and evaluate, then Ainize

FINAL PRODUCT DESCRIPTION (DELIVERABLES)
• a. Final project’s source code
• b. Detailed tutorial of the project and blog entry
• c. Deployed Ainize service URL

FINAL PROJECT SUBMISSION DATE
One month after project start date

1 Like