RescueBox is an open-source, free tool available to all. It is designed to give investigators access to free, machine learning tools that are useful in investigations of child exploitation. We would greatly appreciate your feedback!
By signing up, we will add you to our mailing list and send you updates about new versions of RescueBox as they come out. Please sign up using this form.
You can download the software directly from these links.
Windows software (2.3GB)
The download contains documentation including a product overview, installation notes, and RescueBox usage.
In this release, the machine learning services are as follows. Note that RescueBox is a tool to help investigators work with content, but it's not a forensic tool. In our repository we have reports on accuracy that give a sense of the efficacy of the results.
Face recognition. This service allows users to load images of people into a database. New images can be matched against the database.
Age estimation. This service estimates of the ages of people in images. It also returns the detected gender.
Deep fake detection. This service estimates whether an image is a deep fake. Note that this service is limited and will not work against the newest AI generated images; this cat-and-mouse game will always be true, much like malware scanning.
Audio transcription. This service will return a transcript of words spoken in audio files.
Text summarization. This service will summarize the contents of text files.
Text Summarization. This service will summarize files containing text (.txt/.pdf/.md files) into concise summaries.
In addition, RescueBox can mount UFDR files as if they are drives. This is not a machine learning service but is available from the same menu. Our UFDR mounter is a Python-based FUSE virtual filesystem that allows you to mount .ufdr and .zip archives as read-only directories. This tool lets you browse the contents of forensic archives (like Cellebrite UFDR exports) without extracting them.
You are welcome to simply attach a USB drive of files as well.
This release has been tested on Windows 11 with 64-bit hardware. The software does work if you run it on a machine with only a CPU; try it out with a few images or audio files. However for large-scale work, because the processing is dependent on machine learning models, you won't see good performance unless you run on a machine that has a modern NVIDIA GPU card. Something like 4090 or 5090 or better for large amounts of files.
The software is free of charge. UMass Amherst is research university and a non-profit: our plan to continue development of this open source software is by seeking funding for this work from sponsors. While we seek this funding, we are indeed offering free support for the software for its users. Please ask for our help, offer feedback, and submit bug reports using this form. Again, please note that there is documentation in the windows download.
You can read more about the project from these slides.