Anonymizing AudioVisual Data
This is the official blog post for Anonymization project.
I am very thankful to Google and RedHen Labs for this opportunity. Hoping to get an excellent experience with RedHen Labs.
Timeline Summary
-
18 May: A great day! I will be working with RedHen.
-
1 June: First meeting. Start thinking.
Latest Posts
-
Insights into the internals of rha.py
In this post, I will describe the internal processing of rha.py.
-
Installation of RHA (Red Hen Anonymizer)
In this post, I will summarize the deliverables and the installation procedure.
-
Advanced settings
In this post, I will detail into the advanced settings or the deeper manipulations that are not presented to a general user.
-
Reason for Audio-Video Async
In this post, I will summarize the events that reported the audio-video async and how they were resolved.
-
Using AI generated faces for FSGAN targets
In this post, I will list down some tips for selecting anonymuos faces for anonymization.
-
Results of FSGAN
This post, shows the results of FSGAN.
-
Issues with First-Order-Model (FOM)
Some imperfections were detected with FOM. In this post, I list them down, their possible reasons/solutions and next steps.
-
Results from First-Order-Model (FOM)
Results from first-order-model (FOM).
-
Overview of DeepFake resources
In this post, I will state my approach for exploring Deep Fake.
-
Results of DeepPrivacy
We use DeepPrivacy to anonymize videos.
-
Robust Face Hider and the choices for face detectors
We propose a heuristic to make the face-hider more robust and finally, summarize the face detectors we chose.
-
The output till Eval-1 and meeting summary for 7 July
This post summarizes, the pre-eval-1 meeting and the results at the stage of Eval-1.
-
Rejecting AttGAN, results from Voice anonymization and hiding faces
In this post, I have listed down the status of AttGAN, results from SoX module and results of hiding faces.
-
Applying Current Ideas to General Videos
AttGAN is definitely a good tool for RedHen Transformer. Vid2vid also showed a good example of how a person’s clothes can be changed. What are difficulties of applying them directly...
-
First steps of Anonymizing Audio
The proposal stated 6 effects for anonymization: pitch, bass and treble, distortion, echo, reverb, and wah-wah effect
-
Quantitatively Evaluating the Anonymization from AttGAN
Quantitatively Evaluating the Anonymization from AttGAN
-
AttGAN output for CelebA dataset, some examples
Altering Single Feature
-
Accessing GPUs on CWRU HPC
Refernces
-
Careful Study of Visual Anonymization Papers
Intention
-
Meeting minutes
Start thinking