JustDeepIt

JustDeepIt performs deep learning-based object detection and segmentation of biological images for plant phenotyping and precision agriculture applications.

Key Features:

• Model building and inference modules: Provides modular components for constructing, training, and running deep learning models for image analysis.
• Supported frameworks: Integrates with PyTorch, MMDetection, and Detectron2 for model development and inference.
• Object detection algorithms: Includes implementations of Faster R-CNN, YOLOv3, SSD, and RetinaNet for object detection tasks.
• Instance segmentation: Implements Mask R-CNN for instance-level segmentation of objects such as sugar beets and weeds.
• Salient/semantic segmentation: Uses U²-Net for plant and leaf segmentation tasks.
• API implementation: Exposes model construction and inference functionality via the Python library FastAPI.

Scientific Applications:

• Wheat head detection: Detects wheat heads using Faster R-CNN, YOLOv3, SSD, and RetinaNet.
• Sugar beet and weed segmentation: Segments sugar beets and weeds using Mask R-CNN.
• Plant segmentation: Segments whole plants using U²-Net.
• Leaf segmentation: Performs leaf segmentation using U²-Net.

Methodology:

Implemented using PyTorch, MMDetection, Detectron2, and FastAPI, and supporting Faster R-CNN, YOLOv3, SSD, RetinaNet, Mask R-CNN, and U²-Net for detection and segmentation.