Skip to content

Latest commit

 

History

History
40 lines (24 loc) · 1.34 KB

Readme.md

File metadata and controls

40 lines (24 loc) · 1.34 KB
Raw

Safety Signs detection using machine learning methods

Images

Images found in the thesis are available in the images folder.

Data Generator

Usage

  1. Update data/output/data.yaml with the class labels in order and the number of classes
  2. Put input sign images in data/signs
    1. The first character of the filename MUST be the class index according to the order in data.yaml
  3. Run python generator.py #Images #classes
    1. Default values if not provided : 1000 7
  4. The output data in yolov5 format can be found in data/output

annotator.py can annotate the images in output for visualization.
clear_output.py deletes everything in data/output and data/annotated.

Training Data on Roboflow

https://universe.roboflow.com/signs-ipufk/synth-rcgnr

Validation Data (TGA)

https://universe.roboflow.com/sicherheitskennzeichnung/safety-signs-germany

Training a model

Run the Train YoloV7 Notebook in google colab and follow instructions in the notebook. The training data is downloaded from Roboflow, adjust dataset and api key accordingly.

Validating the performance of a model

Run the Validate YOLOV7 notebook in colab. Right now, the weights for the trained model are downloaded from a hardcoded google drive location. Adjust this for your use case.

The validation dataset is downloaded from Roboflow, again adjust this and the api key.