Amped Authenticate has just received a major upgrade. From advanced video analysis capabilities like the new Coding Tree Units filter and RIFF Viewer, to enhanced deepfake detection with our improved Diffusion Model Deepfake filter (now featuring the Flux model), this release is filled with improvements. You’ll also find expanded redaction options in Annotate’s Hide tool and improved data handling in Frame Analysis, all wrapped in a more intuitive and refined interface.
Dear friends, it’s already time for another shiny, massive update for Amped Authenticate! With this release, we’re enriching the Video Mode with the new Coding Tree Units filter and the RIFF Viewer. We’ve also updated and extended the Diffusion Model Deepfake filter, and added the recent Flux model to the list of known classes. This ensures your deepfake detection capabilities are aligned with the state-of-the-art technology! Keep reading to find out more!
See the New Features in Action
First Things First: New Installer Format
You’ll notice the installer is now available as a ZIP file, which contains a .msi plus many .cab files. This is due to a limitation of MSI files that can’t grow larger than 2 GBs. To install the software, simply unzip the file to a folder and run the MSI installer as you would normally do.
The Coding Tree Units Filter
Amped FIVE users have seen this filter being added to their software a few weeks ago. Now we’re including it in Amped Authenticate as well. Because innovation never stops at Amped, we’ve already improved and enriched it. For a quick introduction to Coding Tree Units (CTUs), you’re invited to check this link and then come back here!
In Amped Authenticate, the Coding Tree Units filter is found in the Compression Analysis category.
Compared to the Amped FIVE version, it features two additions:
- The ability to selectively display the Coding Unit (CU) partitioning and/or the Prediction Unit (PU) partitioning;
- The ability to plot a lot of the information extracted by the filter, allowing for a temporal analysis. In the example below, the plot shows the fixed GOP structure of the analyzed video, reflected by the regularly spaced red spikes, corresponding to frames entirely consisting of intra-coded PUs.
Evaluating and comparing encoding structures is often vital in analyzing file integrity. It is relatively simple to edit and transcode a video file using the same codec, but it is a more complex task to retain the exact parameters. Having the ability to observe these parameters at a CTU level allows users to identify differences to question the authenticity of a file.
For example, when a video is a re-encoded version of a previously compressed stream, its plot may exhibit unusual behavior. For instance, there may be an increase in intra-predicted processing units (PUs) and a decrease in skip PUs in P-frames that correspond to the locations of I-frames in the original video. This phenomenon is similar to the “VPF effect” observed in macroblock-based codecs.
Advanced File Info: New RIFF Viewer Tab and More
The Advanced File Info tool can be accessed from the File Analysis category, by clicking on the dedicated button in the top ribbon, or from the Tools menu.
With this update of Amped Authenticate, there are two important pieces of news.
The New RIFF Viewer Tab
This tab displays the data structure of the Resource Interchange File Format (RIFF), which is used to store data in chunks. It is most commonly used for the AVI Multimedia file type, which is, still today, one of the most common formats for surveillance footage.
The viewer is split into two windows. The left window displays the hex data. The right window displays the RIFF structure.
When you click on any element in the right panel, you’ll be automatically brought to the corresponding file offset in the Hex panel.
The basic structure of a standard AVI RIFF file is:
> RIFF (AVI)
> LIST
> LIST
> LIST
> LIST
> IDX1
Each chunk and sub-chunk has a four-character code denoting the type.
- hdrl: Header List
- avih: AVI Header
- strl: Stream List
- strh: Stream Header
- strf: Stream Format
- movi: Video, Audio and text data
- ..db: Uncompressed Video
- ..dc: Compressed Video
- ..wb: Audio
The flexibility and relative ease of writing data inside the RIFF AVI container does mean that there are many different individual structures specific to the writing application. Therefore, you can use the RIFF Viewer to compare the container structure of a questioned video file against that of a reference file. Remember, you can open multiple instances of Advanced File Info to display them side-by-side.
Learn more about the AVI container by checking out the resources here.
Saving and Loading Data in the Frame Analysis Tab
And now, a long-awaited feature: after moving to the Frame Analysis tab and running the analysis, you can export data to a TSV. Most importantly, you can now load it back into the software GUI. This is a lifesaver for videos with tens of thousands of frames, as the analysis can take some time. You’ll be able to load an existing export from the dialog that appears when activating the Frame Analysis tab.
Updated and Extended Diffusion Model Deepfake Filter
The technology behind deepfakes keeps evolving frantically, so increasingly credible deepfakes are spread on the web and social media.
To help you fight and win the deepfake battle, our developers constantly work to improve Amped Authenticate’s deepfake detection capabilities. Last time, we provided you with the Reflections filter; this time, we’ve empowered the Diffusion Model Deepfake filter.
The Importance of Pixel-level Deepfake Detection
Let me first quickly recap what this filter does: as explained in the related scientific publication, it analyzes the image at the pixel level by extracting a set of features (the CLIP features). Then, uses a machine learning method (a Support Vector Machine) to classify such features as belonging to one of the known AI generators or not. Two remarks are in order:
- This is an AI-based analysis technique, so we recommend using it with a grain of salt. Even when you get a very strong compatibility score, that doesn’t mean you can be totally sure the image is/is not a deepfake. However, it’s a good indicator of more attention being needed. Authenticate gives you several other ways to prove that an image is not authentic.
- When it comes to deepfake detection (and authentication in general), it is very important not to rely solely on the so-called “container” analysis. Container analysis is an excellent way to check an image or video integrity. However, it rapidly becomes useless when the image you have to check comes from social media platforms, since most platforms alter the container and metadata as part of their standard uploading process. Yes, container analysis will inform you it’s not a camera-native image (a rapid check in Amped Authenticate’s File Format will usually do), but not much more than that. This is the reason why, at Amped, we decided to fight the pixel-level detection battle, knowing it’s a tough one.
New Output Class: Flux
With this update, we’ve improved the Diffusion Model Deepfake filter by extending and augmenting the training set, making it more resilient to post-processing. Additionally, we have introduced a new class to detect deepfakes generated with the popular Flux model.
Flux is a recent tool for creating astonishingly realistic diffusion model deepfakes, as shown in the example below.
Most importantly, it can be downloaded and run on reasonably affordable hardware, completely offline. While being very interesting and convenient, this ability also means that those who create content can evade all the filtering and checking usually found in online platforms. That’s why we believed it was very important to add this new model to Amped Authenticate’s deepfake detection arsenal.
New “Inverted” and “Edge Feathering” Modes in Annotate’s Hide
When working with sensitive content that needs careful redaction, it may be necessary to hide everything in the image except for the part of interest. Examples include cases where you want to keep only the face of a subject visible, while redacting the rest of the image to protect others’ privacy or avoid unnecessarily showing indecent content.
The Annotate tool now offers a very convenient way to do that: within the Hide tool, you’ll find a new Invert option. Guess what it does?
Still within the Hide tool, you’ll now be able to use edge feathering to smooth the edges of redaction elements, increasing the visual appeal of your exports.
Other Improvements and Changes
- The “Search Similar Images on the Web” tool has a new behavior. Instead of asking the user permission to upload the image to the web, it will instead open the default browser on the Google Lens page, and open the folder containing the evidence image (or the automatically exported video frame, when used from the Video Mode). Just drag the image into the webpage and the search will begin.
- Increased robustness of Advanced File Info when loading some non-standard video formats
- Made some improvements and fixes to the GUI to make Authenticate prettier than ever!
- Shadows filter: added the warning system so that the filter name will be displayed in red in the filter tree when an unfeasible system is detected.
- JPEG Ghost Plot: improved the plot generation for the report, so that the x-axis only spans values for which the difference has been computed (50-100 by default), improving the visibility.
- Image Mode: improved the ability to correctly load an image even when the file extension does not match the actual image file format.
- File Format filter: we removed the integrity check based on the presence of ICC metadata since they are being added by an increasing number of camera models
Bugfixes
We fixed some minor bugs, and once again, we thank all users who reported them to us!
- Image Mode: Fixed a bug causing the Current System State text not to be colored on some occasions.
- Image Mode: Fixed a delay in updating the viewer after relevant changes in the GUI.
- Image Mode: Fixed text misalignment inside a rotated Text object.
- Image Mode: Fixed an object becoming misaligned when snapping to horizontal or vertical rotation state.
- Video Mode – Motion Vectors: Fixed a crash when data was missing for end frames.
- Video Mode: Fixed a bug that left the Play / Pause button in the wrong state.
- Fixed the auto-resizing of column width in tables, and undesired sorting following it.
Don’t Delay – Update Today
If you have an active support plan, you can update straight away by going into the menu Help>Check for Updates Online within Amped Authenticate. If you need to renew your SMS plan, please contact us or one of our authorized distributors. And remember that you can always manage your license and requests from the Amped Support portal.
