AIDE: Annotation Interface for Data-driven Ecology

AIDE is a modular Web image annotation software suite with machine (deep) learning model support.

Demo

The labeling frontend of AIDE can be tested out in a demo instance here. The following example datasets are currently on display in the demo:

• Image labels on the Snapshot Serengeti camera traps dataset
• Points on the VGG Penguins dataset
• Bounding boxes on the NOAA Arctic Seals aerial imagery
• Semantic segmentation on the Chesapeake Land Cover satellite imagery

Model Marketplace

Many image annotation suites exist, a small number even has some form of machine learning support, often undocumented. AIDE is unlike any of those: it not only integrates over 14 state-of-the-art deep learning models, but allows users to import, fine-tune and share model states through the Model Marketplace. Have you, for example, got a camera traps dataset but no pre-trained model available? AIDE's Model Marketplace likely has you covered! At this moment, the following models are available for download and code-free import into AIDE v2.0 or greater:

Object detection

Dataset	Name	Model type	LR scheduler	mAP	Metrics	AIDE download link
Caltech CT	Faster R-CNN	R50-FPN	3x	0.183	metrics | accuracies	model
Caltech CT	RetinaNet	R50-FPN	2x	0.226	metrics | accuracies	model
Channel Island CT	Faster R-CNN	R50-FPN	3x	0.397	metrics | accuracies	model
Island Conservation CT	Faster R-CNN	R50-FPN	3x	0.435	metrics | accuracies	model
Kuzikus aerial imagery	RetinaNet	R50-FPN	3x	0.389	metrics | accuracies	model

More (and better) models, detailed descriptions, as well as code base for training your own models will be coming soon.

Project Source

Want to know more or use AIDE for your own projects? Good news: AIDE is fully open source, top to bottom!

AIDE GitHub repository