TrafficSigns: [object detection] Add one custom class to YOLOv3 (proof of concept) - Githubissues

tum-phoenix / drive_ml

This is the repository for our Machine Learning applications

13 stars 4 forks source link

TrafficSigns: [object detection] Add one custom class to YOLOv3 (proof of concept) #36

Closed Christiaan1990 closed 3 years ago

Christiaan1990 commented 4 years ago

Goal:

Augment the YOLOv3 algorithm, such that it is able to detect (with 2D bounding box) an additional class

Assigments:

understand the current YOLOv3 algorithm (see repo)
understand which classes are currenty detected
determine the performance with the current classes
add additional data for one additional class
test and evaluate the algorithm
- documentation in the TUM-Wiki: https://wiki.tum.de/display/phoenix/Machine+Learning

Notes:

File: /drive_ml/blob/master/tf_object_detection/Train Yolo3.ipynb
information about the current YOLOv3 implementation and the current datasets: https://wiki.tum.de/display/phoenix