Releases: microsoft/aerial_wildlife_detection
Releases · microsoft/aerial_wildlife_detection
AIDE v2.2
AIDE v2.1
AIDE v2.1
- v2.1 introduces a new installation script for Debian/Ubuntu systems. This has been tested on Ubuntu 20.04 LTS but is still in beta mode. See here.
- Many bug fixes for dependency versions, migration script, UI (new project screen), model import in Model Marketplace
AIDE v2.0
Official release of AIDE version 2.0.
This release contains the following improvements over the previous v1.9:
- New models based on the Detectron2 framework:
- Image classification: AlexNet, DenseNet, MNASNet, MobileNet, ResNet, ResNeXt, ShuffleNet, SqueezeNet, VGG, Wide ResNet
- Object detection: Faster R-CNN, RetinaNet, TridentNet
- Semantic segmentation: DeepLabV3+
- Model Marketplace:
- Share your models across projects with two clicks
- Download trained models to disk, upload to AIDE or import from the Web
- Manual label class assignment for pre-trained models for maximum performance (in addition to dynamic model adaptation)
- Preparation towards code framework for AIDE model zoo (stay tuned for release!)
- Many bug fixes regarding the Workflow Designer, data management, and more
AIDE release for scientific paper
Pre-release of AIDE 2.0 with all functionalities from the paper (Kellenberger, B., Morris, D., Tuia, D.: AIDE: Accelerating Image-Based Ecological Surveys with Interactive Machine Learning. Methods in Ecology and Evolution, in press).
Functionalities contained:
- Labeling interface (image labels, points, bounding boxes, segmentation masks)
- Built-in models (ResNet, RetinaNet, U-Net) with active learning support
- Workflow designer
- Data management
- User and model performance evaluation
Not yet built-in, resp. work in progress:
- Model marketplace
- Improvements to existing functionality
- New, unannounced capabilities