This repository leverages Detectron2, a state-of-the-art object detection and segmentation framework built on PyTorch. Detectron2 provides a flexible and efficient implementation of various algorithms, simplifying tasks like object detection, instance segmentation, and keypoint detection.
The Dataset contains 117 images on "Dental Radiology Scans". 90% images have been taken from the dataset for Training and data-augmentation by 4 ways :
- +90 deg. rotation
- -90 deg. rotation
- horizontal flip
- vertical flip
- +- 1.1 deg. rotation
and total of 269 Training images, 5 Validation images & 22 Test images have been generated from them.
Among them, 100 images have been annotated manually, and the annotated file exported as COCO JSON Format for Object Detection.
All these images used for training Detectron2 and segmented images generated for remaining 169 images. These 169 binary segmented masks are converted into COCO JSON Format for Object Detection annotation.
ALL 269 images and their combined single .json annotation file used for training and 5 validation images their combined single .json annotation file used for validation.
Detectron2 did Object Detection: Identifying and localizing objects with bounding boxes & Semantic Segmentation: Assigning each pixel in an image a class label on the test images.
The Code also save the Binary Predicted Mask of test set.
To streamline your object detection workflow, consider automating tasks with Roboflow and MakeSense.ai:
Roboflow: Easily preprocess, augment, and manage your datasets to improve model performance.
MakeSense.ai: Annotate images with bounding boxes, polygons, keypoints, and more, accelerating the data labeling process.