U-Infuse

U-Infuse enables training custom deep learning object detectors on ecological image data to support detection, classification, and species-distribution analyses from camera trap and related image datasets.

Key Features:

• Custom object detector training: Train object detectors on user-provided image datasets, including camera trap images, for species- and environment-specific models.
• Multiclass and single-class support: Perform both multiclass and single-class training and inference for flexible detection and classification tasks.
• Transfer learning: Leverage transfer learning to initialize and update domain-specific models for faster convergence and reduced training requirements.
• Supervised auto-annotation: Automatically generate annotations for images under supervised control to reduce manual labeling workload.
• Annotation refinement: Refine auto-generated annotations to produce high-quality training datasets.
• Species distribution reports and statistics: Generate species distribution reports and other summary statistics from image-based detections.

Scientific Applications:

• Camera trap image classification: Automated detection and classification of animals in camera trap image datasets.
• Biodiversity monitoring: Derive species occurrence and distribution information from image-derived detections and statistics.
• Ecological research: Develop domain-specific detectors for studies requiring species- and environment-tailored models.

Methodology:

Deep learning-based object detector training with transfer learning; supervised auto-annotation and annotation refinement; support for single-class and multiclass training and inference; generation of species distribution reports and summary statistics.