Deep learning models

Available with Image Analyst license.

The following table provides an overview of the deep learning model types available in AllSource. Each row provides compatible metadata formats and the main use of the specific model type. Where available, accompanying examples are included.

Deep learning model type	Supported metadata	Task	Example
BDCN Edge Detector	Classified tiles	Pixel classification	Land parcel extraction
Change Detector	Classified tiles	Pixel classification (change detection)	Detect building change
ConnectNet	Classified tiles	Pixel classification
CycleGAN	Export tiles CycleGAN	Image translation (unpaired images)	SAR to RGB translation
DeepLab	Classified tiles	Pixel classification
DETReg	PASCAL_VOC_rectangles	Object detection
Faster RCNN	PASCAL_VOC_rectangles KITTI_rectangles	Object detection
Feature classifier	Labeled tiles Imagenet Multilabeled tiles	Object detection	Feature categorization
HED Edge Detector	Classified tiles	Pixel classification	Land parcel extraction
Image Captioner	Image captioning	Image captioning
Mask RCNN	RCNN masks	Object detection (instance segmentation)	Caribou detection and classification
MMDetection	PASCAL_VOC_rectangles KITTI_rectangles	Object detection
MMSegmentation	Classified tiles	Pixel classification
Multi-Task Road Extractor	Classified tiles	Pixel classification	Automatic road extraction
MaX-DeepLab	Panoptic segmentation	Panoptic segmentation
Pix2Pix	Export tiles	Image translation (paired images)	Colorizing historic imagery
PSPNet	Classified tiles	Pixel classification
RetinaNet	PASCAL_VOC_rectangles KITTI_rectangles	Object detection	Utility and vegetation detection
Siam Mask	RCNN Masks	Object tracking
SSD	PASCAL_VOC_rectangles KITTI_rectangles	Object detection	Detect palm tree health
Super-resolution	Superresolution	Image translation (paired images)	Increase image resolution
U-Net	Classified tiles	Pixel classification	Extracting building footprints
YOLOv3	PASCAL_VOC_rectangles KITTI_rectangles	Object detection

Note:

Some of the examples that use the Python Notebook for training can be performed using the Train Deep Learning Model tool.

Deep Learning tasks and tools

Task	Tool
Change detection	Detect Change Using Deep Learning
Image translation (paired and unpaired)	Classify Pixels Using Deep Learning
Object classification	Classify Objects Using Deep Learning
Object detection	Detect Objects Using Deep Learning
Object detection (instance segmentation)	Detect Objects Using Deep Learning
Object tracking	FMV Tracking tab
Pixel classification	Classify Pixels Using Deep Learning