Skip to content

Latest commit

 

History

History
26 lines (20 loc) · 1.52 KB

README.md

File metadata and controls

26 lines (20 loc) · 1.52 KB

FiftyOne-YOLOv5-tutorial + Re-labeling images with Labelbox tool

alt text

A repo containing the starting code/dataset (from roboflow.ai) for visualizing the pred and ground truth together in Voxel-51. This notebook can be used to view predictions with different confidence values which are extracted through trained model .pt file. For visualizing the predictions and ground truth please follow the tutorial in colab notebook by running all cells sequentially. This tutorial can be accessed through the following google colab notebook: https://colab.research.google.com/drive/1xqwIHnOyZBiieCwAYH8Rru7eSmRkzm3-?usp=sharing Support for correcting bounding box mistakes is also added to the colab Notebook

Asumptions (Either one of the following)

  • Having the predictions also as in .txt format used by yolov5.
  • Having the yolov5 trained model weights as .pt file.

New features added (20.06.2022)

  • Integrated metrics such as object mistakeness (missing ground truth boxes based on model predictions) and other available from Fifty-one
  • Integarted LabelBox for relabelling mistakes

Links

Known Erorrs

  • ServiceListenTimeout: fiftyone.core.service.ServerService failed to bind to port 5151
    • Just restart the runtime or try with GPU/TPU based runtime
    • or use older version of fiftyone installtion when working in colab