Confusion Matrix for Object Detection

The ConfusionMatrix class can be used to generate confusion matrix for the object detection task.

Usage

In the test code, you need to declare the ConfusionMatrix class with the appropriate parameters.

conf_mat = ConfusionMatrix(num_classes = 3, CONF_THRESHOLD = 0.3, IOU_THRESHOLD = 0.5)

The class has a function called, process_batch, you can use it update the confusion matrix.

An example usage can be:

conf_mat.process_batch(preds, gt_boxes) where preds are predictions made by model, [N, 6] x1, y1, x2, y2, confidence, class and gt_boxes are the ground truth labels, [M, 4] x1, y1, x2, y2, class.

References

This repository uses a function from Pytorch repository, https://github.com/pytorch/vision/blob/ae228fef1ce176fa3e3949f5db7b6e87a1e33065/torchvision/ops/boxes.py#L138

and

The code is very similar to the repository below and main contribution of this repository is ConfusionMatrix class can be used with all deep learning frameworks.

https://github.com/svpino/tf_object_detection_cm

Citations

If you would like to cite the concept, please use this doi: DOI: 10.5281/zenodo.3724203

Contributions

Thanks to Berkay Ugur Senocak for his contributions for maintenance issues on the repository.