SlideShare a Scribd company logo
1 of 19
Download to read offline
DeepFake Detection: The Importance of Training Data
Preprocessing and Practical Considerations
Dr. Symeon (Akis) Papadopoulos – @sympap
MeVer Team @ Information Technologies Institute (ITI) /
Centre for Research & Technology Hellas (CERTH)
Joint work with Polychronis Charitidis, George Kordopatis-Zilos and
Yiannis Kompatsiaris
AI4Media Workshop on GANs for Media Content Generation, Oct 1, 2020
Media Verification
(MeVer)
DeepFakes
• Content, generated by AI, that seems
authentic to human eye
• Most common form: generation and
manipulation of human face
Source: https://en.wikipedia.org/wiki/Deepfake
Source: https://www.youtube.com/watch?v=iHv6Q9ychnA
Source: Media Forensics and DeepFakes: an overview
Manipulation types
Facial manipulations can
be categorised in four
main different groups:
• Entire face synthesis
• Attribute manipulation
• Identity swap
• Expression swap
Source: DeepFakes and Beyond: A Survey of Face Manipulation and Fake
Detection (Tolosana et al., 2020)
Tolosana, R., et al. (2020). Deepfakes and beyond:
A survey of face manipulation and fake
detection. arXiv preprint arXiv:2001.00179.
Verdoliva, L. (2020). Media forensics and deepfakes:
an overview. arXiv preprint arXiv:2001.06564.
Mirsky, Y., & Lee, W. (2020). The Creation and
Detection of Deepfakes: A Survey. arXiv preprint
arXiv:2004.11138.
WeVerify Project
• WeVerify aims at detecting disinformation in social media and expose
misleading and fabricated content
• Partners: Univ. Sheffield, OntoText, ATC, DW, AFP, EU DisinfoLab, CERTH
• A key outcome is a platform for collaborative content verification,
tracking, and debunking
• Currently, we are developing a deepfake detection service for the
WeVerify platform
• Participation in DeepFake Detection Challenge
https://weverify.eu/
DeepFake Detection Challenge
• Goal: detect videos with facial or voice manipulations
• 2,114 teams participated in the challenge
• Log Loss error evaluation on public and private validation sets
• Public evaluation contained videos with similar transformations as the
training set
• Private evaluation contained organic videos and videos with unknown
transformations from the Internet
• Our final standings:
• public leaderboard: 49 (top 3%) with 0.295 Log Loss error
• private leaderboard: 115 (top 5%) with 0.515 Log Loss error
Source: https://www.kaggle.com/c/deepfake-detection-challenge
DeepFake Detection Challenge - dataset
• Dataset of more than 110k videos
• Approx. 20k REAL and the rest are FAKE
• FAKE videos generated from the REAL
• Models used:
• DeepFake AutoEncoder (DFAE)
• Morphable Mask faceswap (MM/NN)
• Neural Talking Heads (NTH)
• FSGAN
• StyleGAN
Dolhansky, B., Bitton, J., Pflaum, B., Lu, J., Howes, R., Wang,
M., & Ferrer, C. C. (2020). The DeepFake Detection Challenge
Dataset. arXiv preprint arXiv:2006.07397.
Dataset preprocessing - Issues
• Face dataset quality depends on face extraction accuracy (Dlib,
mtcnn, facenet-pytorch, Blazeface)
• Generally all face extraction libraries generate a number of false
positive detections
• Manual tuning can improve the quality of the generated dataset
Deep learning
model
Face
extraction
Frame
extraction
Video
corpus
Noisy data creeping in the training set
• Extracting faces with 1 fps from Kaggle DeepFake Detection Challenge dataset
videos using pytorch implementation of MTCNN face detection
• Observation: False detections are less compared to true detections in a video
Our “noise” filtering approach
• Compute face embeddings for each detected face in video
• Similarity calculation between all face embeddings in a video → similarity graph construction
• Nodes represent faces and two faces are connected if their similarities are greater than 0.8 (solid lines)
• Drop components smaller than N/2 (e.g. component 2)
• N is the number of frames that contain face detections (true or false).
Advantages
• Simple and fast procedure
• No need for manual tuning of the face extraction settings
• Clusters of distinct faces in cases of multiple persons in the video
• This information can be utilized in various ways (e.g. predictions per face)
Faces extracted from multiple video frames
Component 1
Component 2
Experiments
• We trained multiple DeepFake detection models on the DFDC dataset
with and without (baseline) our proposed approach
• Three datasets: a) Celeb-DF, b) FaceForensics++, c) DFDC subset
• For evaluation we examined two aggregation approaches
• avg: prediction is the average of all face predictions
• face: prediction is the max prediction among different avg face predictions
• Results for the EfficientNet-B4 model in terms of Log loss error:
Pre-
processing
CelebDF FaceForensics++ DFDC
avg face avg face avg face
baseline 0,510 0,511 0,563 0,563 0,213 0,198
proposed 0,458 0,456 0,497 0,496 0,195 0,173
Our DFDC Approach - details
• Applied proposed preprocessing approach to clean the generated face dataset
• Face augmentation:
• Horizontal & vertical flip, random crop, rotation, image compression, Gaussian & motion
blurring, brightness, saturation & contrast transformation
• Trained three different models: a) EfficientNet-B3, b) EfficientNet-B4, c) I3D*
• Models trained on face level:
• I3d trained with 10 consecutive face images exploiting temporal information.
• EfficientNet models trained on single face images
• Per model:
• Added two dense layers with dropout after the backbone architecture with 256 and 1 units
• Used the sigmoid activation for the last layer
* ignoring the optical flow stream
Our DFDC approach – inference
pre-processing model inference post-processing
Lessons from other DFDC teams
• Most approaches ensemble multiple EfficientNet architectures (B3-B7) and
some of them were trained on different seeds
• ResNeXT was another architecture used by a top-performing solutions
combined with 3D architectures such as I3D, 3D ResNet34, MC3 & R2+1D
• Several approaches increased the margin of the detected facial bounding
box to further improve results.
• We used an additional margin of 20% but other works proposed a higher proportion.
• To improve generalization:
• Domain-specific augmentations: a) half face removal horizontally or vertically, b)
landmark (eyes, nose, or mouth) removal
• Mixup augmentations
Practical challenges
• Limited generalization
• This observation applies to most submissions. The winning team scored
0.20336 in public validation and only 0.42798 in the private (Log Loss)
• Overfitting
• The best submission in the public leaderboard scored 0.19207 but in the
private evaluation the error was 0.57468, leading to the 904-th position!
• Broad problem scope
• The term DeepFake may refer to every possible manipulation and generation
• Constantly increasing manipulation and generation techniques
• A detector is only trained with a subset of these manipulations
DeepFake Detection in the Wild
• Videos in the wild usually contain multiple scenes
• Only a subset of these scenes may contain DeepFakes
• Detection process might be slow for multi-shot videos (even short ones)
• Low quality videos
• Low quality faces tend to fool classifiers
• Small detected and fast-moving faces
• Usually lead to noisy predictions
• Changes in the environment
• Moving obstacles in front of the faces
• Changes in lighting
DeepFake Detection Service @ WeVerify
https://www.youtube.com/watch?v=cVljNV
V5VPw&ab_channel=TheFakening
More details at TTO 2020
Charitidis, P., Kordopatis-Zilos, G., Papadopoulos, S., & Kompatsiaris, Y.
(2020). Investigating the impact of preprocessing and prediction
aggregation on the DeepFake detection task. Proceedings of the
Conference for Truth and Trust Online (TTO) [to appear],
https://arxiv.org/abs/2006.07084
https://truthandtrustonline.com/
Thank you!
Dr. Symeon Papadopoulos
papadop@iti.gr
@sympap
Media Verification (MeVer)
https://mever.iti.gr/
@meverteam https://ai4media.eu/
https://weverify.eu/

More Related Content

What's hot

A Proposed Framework for Robust Face Identification System
A Proposed Framework for Robust Face Identification SystemA Proposed Framework for Robust Face Identification System
A Proposed Framework for Robust Face Identification SystemAhmed Gad
 
MixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing
MixTaiwan 20170222 清大電機 孫民 AI The Next Big ThingMixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing
MixTaiwan 20170222 清大電機 孫民 AI The Next Big ThingMix Taiwan
 
Design of a Robotic Guide/Usher
Design of a Robotic Guide/UsherDesign of a Robotic Guide/Usher
Design of a Robotic Guide/UsherTyler Aaron
 
Verifying Multimedia Content on the Internet
Verifying Multimedia Content on the InternetVerifying Multimedia Content on the Internet
Verifying Multimedia Content on the InternetSymeon Papadopoulos
 
Mobile Interactive Hologram Verification
Mobile Interactive Hologram VerificationMobile Interactive Hologram Verification
Mobile Interactive Hologram VerificationJens Grubert
 
Face Detection - David
Face Detection - DavidFace Detection - David
Face Detection - DavidVu Tran
 

What's hot (7)

A Proposed Framework for Robust Face Identification System
A Proposed Framework for Robust Face Identification SystemA Proposed Framework for Robust Face Identification System
A Proposed Framework for Robust Face Identification System
 
MixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing
MixTaiwan 20170222 清大電機 孫民 AI The Next Big ThingMixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing
MixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing
 
Face Recognition - Elisa Sayrol - UPC Barcelona 2018
Face Recognition - Elisa Sayrol - UPC Barcelona 2018Face Recognition - Elisa Sayrol - UPC Barcelona 2018
Face Recognition - Elisa Sayrol - UPC Barcelona 2018
 
Design of a Robotic Guide/Usher
Design of a Robotic Guide/UsherDesign of a Robotic Guide/Usher
Design of a Robotic Guide/Usher
 
Verifying Multimedia Content on the Internet
Verifying Multimedia Content on the InternetVerifying Multimedia Content on the Internet
Verifying Multimedia Content on the Internet
 
Mobile Interactive Hologram Verification
Mobile Interactive Hologram VerificationMobile Interactive Hologram Verification
Mobile Interactive Hologram Verification
 
Face Detection - David
Face Detection - DavidFace Detection - David
Face Detection - David
 

Similar to Deepfake Detection: The Importance of Training Data Preprocessing and Practical Considerations

DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...
DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...
DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...Symeon Papadopoulos
 
Face Recognition System for Door Unlocking
Face Recognition System for Door UnlockingFace Recognition System for Door Unlocking
Face Recognition System for Door UnlockingHassan Tariq
 
Face detection and recognition
Face detection and recognitionFace detection and recognition
Face detection and recognitionPankaj Thakur
 
Biometric Recognition using Deep Learning
Biometric Recognition using Deep LearningBiometric Recognition using Deep Learning
Biometric Recognition using Deep LearningSahithiKotha2
 
VOGIN-IP-lezing-Zeno_ geradts
VOGIN-IP-lezing-Zeno_ geradtsVOGIN-IP-lezing-Zeno_ geradts
VOGIN-IP-lezing-Zeno_ geradtsvoginip
 
Automated_attendance_system_project.pptx
Automated_attendance_system_project.pptxAutomated_attendance_system_project.pptx
Automated_attendance_system_project.pptxNaveensai51
 
Technical Workshop - Win32/Georbot Analysis
Technical Workshop - Win32/Georbot AnalysisTechnical Workshop - Win32/Georbot Analysis
Technical Workshop - Win32/Georbot AnalysisPositive Hack Days
 
DeepFake: Trick or Treat
DeepFake: Trick or TreatDeepFake: Trick or Treat
DeepFake: Trick or TreatSamama Khan
 
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde..."Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...Edge AI and Vision Alliance
 
Attendance System using Facial Recognition
Attendance System using Facial RecognitionAttendance System using Facial Recognition
Attendance System using Facial RecognitionIRJET Journal
 
Human Face Identification
Human Face IdentificationHuman Face Identification
Human Face Identificationbhupesh lahare
 
Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018InVID Project
 
Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...
Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...
Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...Vienna Data Science Group
 
FASSOLD Deep learning for semantic analysis and annotation of conventional an...
FASSOLD Deep learning for semantic analysis and annotation of conventional an...FASSOLD Deep learning for semantic analysis and annotation of conventional an...
FASSOLD Deep learning for semantic analysis and annotation of conventional an...FIAT/IFTA
 
Recent Advances in Face Analysis: database, methods, and software.
Recent Advances in Face Analysis: database, methods, and software.Recent Advances in Face Analysis: database, methods, and software.
Recent Advances in Face Analysis: database, methods, and software.Taowei Huang
 
Face recognition v1
Face recognition v1Face recognition v1
Face recognition v1San Kim
 
IRJET- Real Time Attendance System using Face Recognition
IRJET-  	  Real Time Attendance System using Face RecognitionIRJET-  	  Real Time Attendance System using Face Recognition
IRJET- Real Time Attendance System using Face RecognitionIRJET Journal
 

Similar to Deepfake Detection: The Importance of Training Data Preprocessing and Practical Considerations (20)

DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...
DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...
DeepFake Detection: Challenges, Progress and Hands-on Demonstration of Techno...
 
Deep fakes and beyond
Deep fakes and beyondDeep fakes and beyond
Deep fakes and beyond
 
Face Recognition System for Door Unlocking
Face Recognition System for Door UnlockingFace Recognition System for Door Unlocking
Face Recognition System for Door Unlocking
 
Face detection and recognition
Face detection and recognitionFace detection and recognition
Face detection and recognition
 
Biometric Recognition using Deep Learning
Biometric Recognition using Deep LearningBiometric Recognition using Deep Learning
Biometric Recognition using Deep Learning
 
VOGIN-IP-lezing-Zeno_ geradts
VOGIN-IP-lezing-Zeno_ geradtsVOGIN-IP-lezing-Zeno_ geradts
VOGIN-IP-lezing-Zeno_ geradts
 
A guide to Face Detection in Python.pdf
A guide to Face Detection in Python.pdfA guide to Face Detection in Python.pdf
A guide to Face Detection in Python.pdf
 
Automated_attendance_system_project.pptx
Automated_attendance_system_project.pptxAutomated_attendance_system_project.pptx
Automated_attendance_system_project.pptx
 
Technical Workshop - Win32/Georbot Analysis
Technical Workshop - Win32/Georbot AnalysisTechnical Workshop - Win32/Georbot Analysis
Technical Workshop - Win32/Georbot Analysis
 
DeepFake: Trick or Treat
DeepFake: Trick or TreatDeepFake: Trick or Treat
DeepFake: Trick or Treat
 
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde..."Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
 
Attendance System using Facial Recognition
Attendance System using Facial RecognitionAttendance System using Facial Recognition
Attendance System using Facial Recognition
 
Human Face Identification
Human Face IdentificationHuman Face Identification
Human Face Identification
 
Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018Presentation of the InVID verification technologies at IPTC 2018
Presentation of the InVID verification technologies at IPTC 2018
 
Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...
Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...
Wastian, Brunmeir - Data Analyses in Industrial Applications: From Predictive...
 
FASSOLD Deep learning for semantic analysis and annotation of conventional an...
FASSOLD Deep learning for semantic analysis and annotation of conventional an...FASSOLD Deep learning for semantic analysis and annotation of conventional an...
FASSOLD Deep learning for semantic analysis and annotation of conventional an...
 
Recent Advances in Face Analysis: database, methods, and software.
Recent Advances in Face Analysis: database, methods, and software.Recent Advances in Face Analysis: database, methods, and software.
Recent Advances in Face Analysis: database, methods, and software.
 
Face recognition v1
Face recognition v1Face recognition v1
Face recognition v1
 
ESE presentation.pptx
ESE presentation.pptxESE presentation.pptx
ESE presentation.pptx
 
IRJET- Real Time Attendance System using Face Recognition
IRJET-  	  Real Time Attendance System using Face RecognitionIRJET-  	  Real Time Attendance System using Face Recognition
IRJET- Real Time Attendance System using Face Recognition
 

More from Symeon Papadopoulos

Deepfakes: An Emerging Internet Threat and their Detection
Deepfakes: An Emerging Internet Threat and their DetectionDeepfakes: An Emerging Internet Threat and their Detection
Deepfakes: An Emerging Internet Threat and their DetectionSymeon Papadopoulos
 
Knowledge-based Fusion for Image Tampering Localization
Knowledge-based Fusion for Image Tampering LocalizationKnowledge-based Fusion for Image Tampering Localization
Knowledge-based Fusion for Image Tampering LocalizationSymeon Papadopoulos
 
COVID-19 Infodemic vs Contact Tracing
COVID-19 Infodemic vs Contact TracingCOVID-19 Infodemic vs Contact Tracing
COVID-19 Infodemic vs Contact TracingSymeon Papadopoulos
 
Similarity-based retrieval of multimedia content
Similarity-based retrieval of multimedia contentSimilarity-based retrieval of multimedia content
Similarity-based retrieval of multimedia contentSymeon Papadopoulos
 
Twitter-based Sensing of City-level Air Quality
Twitter-based Sensing of City-level Air QualityTwitter-based Sensing of City-level Air Quality
Twitter-based Sensing of City-level Air QualitySymeon Papadopoulos
 
Aggregating and Analyzing the Context of Social Media Content
Aggregating and Analyzing the Context of Social Media ContentAggregating and Analyzing the Context of Social Media Content
Aggregating and Analyzing the Context of Social Media ContentSymeon Papadopoulos
 
Learning to detect Misleading Content on Twitter
Learning to detect Misleading Content on TwitterLearning to detect Misleading Content on Twitter
Learning to detect Misleading Content on TwitterSymeon Papadopoulos
 
Near-Duplicate Video Retrieval by Aggregating Intermediate CNN Layers
Near-Duplicate Video Retrieval by Aggregating Intermediate CNN LayersNear-Duplicate Video Retrieval by Aggregating Intermediate CNN Layers
Near-Duplicate Video Retrieval by Aggregating Intermediate CNN LayersSymeon Papadopoulos
 
Verifying Multimedia Use at MediaEval 2016
Verifying Multimedia Use at MediaEval 2016Verifying Multimedia Use at MediaEval 2016
Verifying Multimedia Use at MediaEval 2016Symeon Papadopoulos
 
Placing Images with Refined Language Models and Similarity Search with PCA-re...
Placing Images with Refined Language Models and Similarity Search with PCA-re...Placing Images with Refined Language Models and Similarity Search with PCA-re...
Placing Images with Refined Language Models and Similarity Search with PCA-re...Symeon Papadopoulos
 
In-depth Exploration of Geotagging Performance
In-depth Exploration of Geotagging PerformanceIn-depth Exploration of Geotagging Performance
In-depth Exploration of Geotagging PerformanceSymeon Papadopoulos
 
Perceived versus Actual Predictability of Personal Information in Social Netw...
Perceived versus Actual Predictability of Personal Information in Social Netw...Perceived versus Actual Predictability of Personal Information in Social Netw...
Perceived versus Actual Predictability of Personal Information in Social Netw...Symeon Papadopoulos
 
Web and Social Media Image Forensics for News Professionals
Web and Social Media Image Forensics for News ProfessionalsWeb and Social Media Image Forensics for News Professionals
Web and Social Media Image Forensics for News ProfessionalsSymeon Papadopoulos
 
Predicting News Popularity by Mining Online Discussions
Predicting News Popularity by Mining Online DiscussionsPredicting News Popularity by Mining Online Discussions
Predicting News Popularity by Mining Online DiscussionsSymeon Papadopoulos
 
Finding Diverse Social Images at MediaEval 2015
Finding Diverse Social Images at MediaEval 2015Finding Diverse Social Images at MediaEval 2015
Finding Diverse Social Images at MediaEval 2015Symeon Papadopoulos
 
CERTH/CEA LIST at MediaEval Placing Task 2015
CERTH/CEA LIST at MediaEval Placing Task 2015CERTH/CEA LIST at MediaEval Placing Task 2015
CERTH/CEA LIST at MediaEval Placing Task 2015Symeon Papadopoulos
 
Verifying Multimedia Use at MediaEval 2015
Verifying Multimedia Use at MediaEval 2015Verifying Multimedia Use at MediaEval 2015
Verifying Multimedia Use at MediaEval 2015Symeon Papadopoulos
 
Detecting image splicing in the wild Web
Detecting image splicing in the wild WebDetecting image splicing in the wild Web
Detecting image splicing in the wild WebSymeon Papadopoulos
 
Learning to Classify Users in Online Interaction Networks
Learning to Classify Users in Online Interaction NetworksLearning to Classify Users in Online Interaction Networks
Learning to Classify Users in Online Interaction NetworksSymeon Papadopoulos
 

More from Symeon Papadopoulos (20)

Deepfakes: An Emerging Internet Threat and their Detection
Deepfakes: An Emerging Internet Threat and their DetectionDeepfakes: An Emerging Internet Threat and their Detection
Deepfakes: An Emerging Internet Threat and their Detection
 
Knowledge-based Fusion for Image Tampering Localization
Knowledge-based Fusion for Image Tampering LocalizationKnowledge-based Fusion for Image Tampering Localization
Knowledge-based Fusion for Image Tampering Localization
 
COVID-19 Infodemic vs Contact Tracing
COVID-19 Infodemic vs Contact TracingCOVID-19 Infodemic vs Contact Tracing
COVID-19 Infodemic vs Contact Tracing
 
Similarity-based retrieval of multimedia content
Similarity-based retrieval of multimedia contentSimilarity-based retrieval of multimedia content
Similarity-based retrieval of multimedia content
 
Twitter-based Sensing of City-level Air Quality
Twitter-based Sensing of City-level Air QualityTwitter-based Sensing of City-level Air Quality
Twitter-based Sensing of City-level Air Quality
 
Aggregating and Analyzing the Context of Social Media Content
Aggregating and Analyzing the Context of Social Media ContentAggregating and Analyzing the Context of Social Media Content
Aggregating and Analyzing the Context of Social Media Content
 
Learning to detect Misleading Content on Twitter
Learning to detect Misleading Content on TwitterLearning to detect Misleading Content on Twitter
Learning to detect Misleading Content on Twitter
 
Near-Duplicate Video Retrieval by Aggregating Intermediate CNN Layers
Near-Duplicate Video Retrieval by Aggregating Intermediate CNN LayersNear-Duplicate Video Retrieval by Aggregating Intermediate CNN Layers
Near-Duplicate Video Retrieval by Aggregating Intermediate CNN Layers
 
Verifying Multimedia Use at MediaEval 2016
Verifying Multimedia Use at MediaEval 2016Verifying Multimedia Use at MediaEval 2016
Verifying Multimedia Use at MediaEval 2016
 
Multimedia Privacy
Multimedia PrivacyMultimedia Privacy
Multimedia Privacy
 
Placing Images with Refined Language Models and Similarity Search with PCA-re...
Placing Images with Refined Language Models and Similarity Search with PCA-re...Placing Images with Refined Language Models and Similarity Search with PCA-re...
Placing Images with Refined Language Models and Similarity Search with PCA-re...
 
In-depth Exploration of Geotagging Performance
In-depth Exploration of Geotagging PerformanceIn-depth Exploration of Geotagging Performance
In-depth Exploration of Geotagging Performance
 
Perceived versus Actual Predictability of Personal Information in Social Netw...
Perceived versus Actual Predictability of Personal Information in Social Netw...Perceived versus Actual Predictability of Personal Information in Social Netw...
Perceived versus Actual Predictability of Personal Information in Social Netw...
 
Web and Social Media Image Forensics for News Professionals
Web and Social Media Image Forensics for News ProfessionalsWeb and Social Media Image Forensics for News Professionals
Web and Social Media Image Forensics for News Professionals
 
Predicting News Popularity by Mining Online Discussions
Predicting News Popularity by Mining Online DiscussionsPredicting News Popularity by Mining Online Discussions
Predicting News Popularity by Mining Online Discussions
 
Finding Diverse Social Images at MediaEval 2015
Finding Diverse Social Images at MediaEval 2015Finding Diverse Social Images at MediaEval 2015
Finding Diverse Social Images at MediaEval 2015
 
CERTH/CEA LIST at MediaEval Placing Task 2015
CERTH/CEA LIST at MediaEval Placing Task 2015CERTH/CEA LIST at MediaEval Placing Task 2015
CERTH/CEA LIST at MediaEval Placing Task 2015
 
Verifying Multimedia Use at MediaEval 2015
Verifying Multimedia Use at MediaEval 2015Verifying Multimedia Use at MediaEval 2015
Verifying Multimedia Use at MediaEval 2015
 
Detecting image splicing in the wild Web
Detecting image splicing in the wild WebDetecting image splicing in the wild Web
Detecting image splicing in the wild Web
 
Learning to Classify Users in Online Interaction Networks
Learning to Classify Users in Online Interaction NetworksLearning to Classify Users in Online Interaction Networks
Learning to Classify Users in Online Interaction Networks
 

Recently uploaded

Dev Dives: Streamline document processing with UiPath Studio Web
Dev Dives: Streamline document processing with UiPath Studio WebDev Dives: Streamline document processing with UiPath Studio Web
Dev Dives: Streamline document processing with UiPath Studio WebUiPathCommunity
 
Developer Data Modeling Mistakes: From Postgres to NoSQL
Developer Data Modeling Mistakes: From Postgres to NoSQLDeveloper Data Modeling Mistakes: From Postgres to NoSQL
Developer Data Modeling Mistakes: From Postgres to NoSQLScyllaDB
 
TrustArc Webinar - How to Build Consumer Trust Through Data Privacy
TrustArc Webinar - How to Build Consumer Trust Through Data PrivacyTrustArc Webinar - How to Build Consumer Trust Through Data Privacy
TrustArc Webinar - How to Build Consumer Trust Through Data PrivacyTrustArc
 
Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024BookNet Canada
 
Anypoint Exchange: It’s Not Just a Repo!
Anypoint Exchange: It’s Not Just a Repo!Anypoint Exchange: It’s Not Just a Repo!
Anypoint Exchange: It’s Not Just a Repo!Manik S Magar
 
The Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptx
The Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptxThe Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptx
The Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptxLoriGlavin3
 
Sample pptx for embedding into website for demo
Sample pptx for embedding into website for demoSample pptx for embedding into website for demo
Sample pptx for embedding into website for demoHarshalMandlekar2
 
DevoxxFR 2024 Reproducible Builds with Apache Maven
DevoxxFR 2024 Reproducible Builds with Apache MavenDevoxxFR 2024 Reproducible Builds with Apache Maven
DevoxxFR 2024 Reproducible Builds with Apache MavenHervé Boutemy
 
"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek SchlawackFwdays
 
What is Artificial Intelligence?????????
What is Artificial Intelligence?????????What is Artificial Intelligence?????????
What is Artificial Intelligence?????????blackmambaettijean
 
How to write a Business Continuity Plan
How to write a Business Continuity PlanHow to write a Business Continuity Plan
How to write a Business Continuity PlanDatabarracks
 
Unraveling Multimodality with Large Language Models.pdf
Unraveling Multimodality with Large Language Models.pdfUnraveling Multimodality with Large Language Models.pdf
Unraveling Multimodality with Large Language Models.pdfAlex Barbosa Coqueiro
 
TeamStation AI System Report LATAM IT Salaries 2024
TeamStation AI System Report LATAM IT Salaries 2024TeamStation AI System Report LATAM IT Salaries 2024
TeamStation AI System Report LATAM IT Salaries 2024Lonnie McRorey
 
Gen AI in Business - Global Trends Report 2024.pdf
Gen AI in Business - Global Trends Report 2024.pdfGen AI in Business - Global Trends Report 2024.pdf
Gen AI in Business - Global Trends Report 2024.pdfAddepto
 
Moving Beyond Passwords: FIDO Paris Seminar.pdf
Moving Beyond Passwords: FIDO Paris Seminar.pdfMoving Beyond Passwords: FIDO Paris Seminar.pdf
Moving Beyond Passwords: FIDO Paris Seminar.pdfLoriGlavin3
 
From Family Reminiscence to Scholarly Archive .
From Family Reminiscence to Scholarly Archive .From Family Reminiscence to Scholarly Archive .
From Family Reminiscence to Scholarly Archive .Alan Dix
 
"Debugging python applications inside k8s environment", Andrii Soldatenko
"Debugging python applications inside k8s environment", Andrii Soldatenko"Debugging python applications inside k8s environment", Andrii Soldatenko
"Debugging python applications inside k8s environment", Andrii SoldatenkoFwdays
 
A Journey Into the Emotions of Software Developers
A Journey Into the Emotions of Software DevelopersA Journey Into the Emotions of Software Developers
A Journey Into the Emotions of Software DevelopersNicole Novielli
 
The Ultimate Guide to Choosing WordPress Pros and Cons
The Ultimate Guide to Choosing WordPress Pros and ConsThe Ultimate Guide to Choosing WordPress Pros and Cons
The Ultimate Guide to Choosing WordPress Pros and ConsPixlogix Infotech
 
unit 4 immunoblotting technique complete.pptx
unit 4 immunoblotting technique complete.pptxunit 4 immunoblotting technique complete.pptx
unit 4 immunoblotting technique complete.pptxBkGupta21
 

Recently uploaded (20)

Dev Dives: Streamline document processing with UiPath Studio Web
Dev Dives: Streamline document processing with UiPath Studio WebDev Dives: Streamline document processing with UiPath Studio Web
Dev Dives: Streamline document processing with UiPath Studio Web
 
Developer Data Modeling Mistakes: From Postgres to NoSQL
Developer Data Modeling Mistakes: From Postgres to NoSQLDeveloper Data Modeling Mistakes: From Postgres to NoSQL
Developer Data Modeling Mistakes: From Postgres to NoSQL
 
TrustArc Webinar - How to Build Consumer Trust Through Data Privacy
TrustArc Webinar - How to Build Consumer Trust Through Data PrivacyTrustArc Webinar - How to Build Consumer Trust Through Data Privacy
TrustArc Webinar - How to Build Consumer Trust Through Data Privacy
 
Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024
Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024
 
Anypoint Exchange: It’s Not Just a Repo!
Anypoint Exchange: It’s Not Just a Repo!Anypoint Exchange: It’s Not Just a Repo!
Anypoint Exchange: It’s Not Just a Repo!
 
The Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptx
The Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptxThe Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptx
The Role of FIDO in a Cyber Secure Netherlands: FIDO Paris Seminar.pptx
 
Sample pptx for embedding into website for demo
Sample pptx for embedding into website for demoSample pptx for embedding into website for demo
Sample pptx for embedding into website for demo
 
DevoxxFR 2024 Reproducible Builds with Apache Maven
DevoxxFR 2024 Reproducible Builds with Apache MavenDevoxxFR 2024 Reproducible Builds with Apache Maven
DevoxxFR 2024 Reproducible Builds with Apache Maven
 
"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
"Subclassing and Composition – A Pythonic Tour of Trade-Offs", Hynek Schlawack
 
What is Artificial Intelligence?????????
What is Artificial Intelligence?????????What is Artificial Intelligence?????????
What is Artificial Intelligence?????????
 
How to write a Business Continuity Plan
How to write a Business Continuity PlanHow to write a Business Continuity Plan
How to write a Business Continuity Plan
 
Unraveling Multimodality with Large Language Models.pdf
Unraveling Multimodality with Large Language Models.pdfUnraveling Multimodality with Large Language Models.pdf
Unraveling Multimodality with Large Language Models.pdf
 
TeamStation AI System Report LATAM IT Salaries 2024
TeamStation AI System Report LATAM IT Salaries 2024TeamStation AI System Report LATAM IT Salaries 2024
TeamStation AI System Report LATAM IT Salaries 2024
 
Gen AI in Business - Global Trends Report 2024.pdf
Gen AI in Business - Global Trends Report 2024.pdfGen AI in Business - Global Trends Report 2024.pdf
Gen AI in Business - Global Trends Report 2024.pdf
 
Moving Beyond Passwords: FIDO Paris Seminar.pdf
Moving Beyond Passwords: FIDO Paris Seminar.pdfMoving Beyond Passwords: FIDO Paris Seminar.pdf
Moving Beyond Passwords: FIDO Paris Seminar.pdf
 
From Family Reminiscence to Scholarly Archive .
From Family Reminiscence to Scholarly Archive .From Family Reminiscence to Scholarly Archive .
From Family Reminiscence to Scholarly Archive .
 
"Debugging python applications inside k8s environment", Andrii Soldatenko
"Debugging python applications inside k8s environment", Andrii Soldatenko"Debugging python applications inside k8s environment", Andrii Soldatenko
"Debugging python applications inside k8s environment", Andrii Soldatenko
 
A Journey Into the Emotions of Software Developers
A Journey Into the Emotions of Software DevelopersA Journey Into the Emotions of Software Developers
A Journey Into the Emotions of Software Developers
 
The Ultimate Guide to Choosing WordPress Pros and Cons
The Ultimate Guide to Choosing WordPress Pros and ConsThe Ultimate Guide to Choosing WordPress Pros and Cons
The Ultimate Guide to Choosing WordPress Pros and Cons
 
unit 4 immunoblotting technique complete.pptx
unit 4 immunoblotting technique complete.pptxunit 4 immunoblotting technique complete.pptx
unit 4 immunoblotting technique complete.pptx
 

Deepfake Detection: The Importance of Training Data Preprocessing and Practical Considerations

  • 1. DeepFake Detection: The Importance of Training Data Preprocessing and Practical Considerations Dr. Symeon (Akis) Papadopoulos – @sympap MeVer Team @ Information Technologies Institute (ITI) / Centre for Research & Technology Hellas (CERTH) Joint work with Polychronis Charitidis, George Kordopatis-Zilos and Yiannis Kompatsiaris AI4Media Workshop on GANs for Media Content Generation, Oct 1, 2020 Media Verification (MeVer)
  • 2. DeepFakes • Content, generated by AI, that seems authentic to human eye • Most common form: generation and manipulation of human face Source: https://en.wikipedia.org/wiki/Deepfake Source: https://www.youtube.com/watch?v=iHv6Q9ychnA Source: Media Forensics and DeepFakes: an overview
  • 3. Manipulation types Facial manipulations can be categorised in four main different groups: • Entire face synthesis • Attribute manipulation • Identity swap • Expression swap Source: DeepFakes and Beyond: A Survey of Face Manipulation and Fake Detection (Tolosana et al., 2020) Tolosana, R., et al. (2020). Deepfakes and beyond: A survey of face manipulation and fake detection. arXiv preprint arXiv:2001.00179. Verdoliva, L. (2020). Media forensics and deepfakes: an overview. arXiv preprint arXiv:2001.06564. Mirsky, Y., & Lee, W. (2020). The Creation and Detection of Deepfakes: A Survey. arXiv preprint arXiv:2004.11138.
  • 4. WeVerify Project • WeVerify aims at detecting disinformation in social media and expose misleading and fabricated content • Partners: Univ. Sheffield, OntoText, ATC, DW, AFP, EU DisinfoLab, CERTH • A key outcome is a platform for collaborative content verification, tracking, and debunking • Currently, we are developing a deepfake detection service for the WeVerify platform • Participation in DeepFake Detection Challenge https://weverify.eu/
  • 5. DeepFake Detection Challenge • Goal: detect videos with facial or voice manipulations • 2,114 teams participated in the challenge • Log Loss error evaluation on public and private validation sets • Public evaluation contained videos with similar transformations as the training set • Private evaluation contained organic videos and videos with unknown transformations from the Internet • Our final standings: • public leaderboard: 49 (top 3%) with 0.295 Log Loss error • private leaderboard: 115 (top 5%) with 0.515 Log Loss error Source: https://www.kaggle.com/c/deepfake-detection-challenge
  • 6. DeepFake Detection Challenge - dataset • Dataset of more than 110k videos • Approx. 20k REAL and the rest are FAKE • FAKE videos generated from the REAL • Models used: • DeepFake AutoEncoder (DFAE) • Morphable Mask faceswap (MM/NN) • Neural Talking Heads (NTH) • FSGAN • StyleGAN Dolhansky, B., Bitton, J., Pflaum, B., Lu, J., Howes, R., Wang, M., & Ferrer, C. C. (2020). The DeepFake Detection Challenge Dataset. arXiv preprint arXiv:2006.07397.
  • 7. Dataset preprocessing - Issues • Face dataset quality depends on face extraction accuracy (Dlib, mtcnn, facenet-pytorch, Blazeface) • Generally all face extraction libraries generate a number of false positive detections • Manual tuning can improve the quality of the generated dataset Deep learning model Face extraction Frame extraction Video corpus
  • 8. Noisy data creeping in the training set • Extracting faces with 1 fps from Kaggle DeepFake Detection Challenge dataset videos using pytorch implementation of MTCNN face detection • Observation: False detections are less compared to true detections in a video
  • 9. Our “noise” filtering approach • Compute face embeddings for each detected face in video • Similarity calculation between all face embeddings in a video → similarity graph construction • Nodes represent faces and two faces are connected if their similarities are greater than 0.8 (solid lines) • Drop components smaller than N/2 (e.g. component 2) • N is the number of frames that contain face detections (true or false).
  • 10. Advantages • Simple and fast procedure • No need for manual tuning of the face extraction settings • Clusters of distinct faces in cases of multiple persons in the video • This information can be utilized in various ways (e.g. predictions per face) Faces extracted from multiple video frames Component 1 Component 2
  • 11. Experiments • We trained multiple DeepFake detection models on the DFDC dataset with and without (baseline) our proposed approach • Three datasets: a) Celeb-DF, b) FaceForensics++, c) DFDC subset • For evaluation we examined two aggregation approaches • avg: prediction is the average of all face predictions • face: prediction is the max prediction among different avg face predictions • Results for the EfficientNet-B4 model in terms of Log loss error: Pre- processing CelebDF FaceForensics++ DFDC avg face avg face avg face baseline 0,510 0,511 0,563 0,563 0,213 0,198 proposed 0,458 0,456 0,497 0,496 0,195 0,173
  • 12. Our DFDC Approach - details • Applied proposed preprocessing approach to clean the generated face dataset • Face augmentation: • Horizontal & vertical flip, random crop, rotation, image compression, Gaussian & motion blurring, brightness, saturation & contrast transformation • Trained three different models: a) EfficientNet-B3, b) EfficientNet-B4, c) I3D* • Models trained on face level: • I3d trained with 10 consecutive face images exploiting temporal information. • EfficientNet models trained on single face images • Per model: • Added two dense layers with dropout after the backbone architecture with 256 and 1 units • Used the sigmoid activation for the last layer * ignoring the optical flow stream
  • 13. Our DFDC approach – inference pre-processing model inference post-processing
  • 14. Lessons from other DFDC teams • Most approaches ensemble multiple EfficientNet architectures (B3-B7) and some of them were trained on different seeds • ResNeXT was another architecture used by a top-performing solutions combined with 3D architectures such as I3D, 3D ResNet34, MC3 & R2+1D • Several approaches increased the margin of the detected facial bounding box to further improve results. • We used an additional margin of 20% but other works proposed a higher proportion. • To improve generalization: • Domain-specific augmentations: a) half face removal horizontally or vertically, b) landmark (eyes, nose, or mouth) removal • Mixup augmentations
  • 15. Practical challenges • Limited generalization • This observation applies to most submissions. The winning team scored 0.20336 in public validation and only 0.42798 in the private (Log Loss) • Overfitting • The best submission in the public leaderboard scored 0.19207 but in the private evaluation the error was 0.57468, leading to the 904-th position! • Broad problem scope • The term DeepFake may refer to every possible manipulation and generation • Constantly increasing manipulation and generation techniques • A detector is only trained with a subset of these manipulations
  • 16. DeepFake Detection in the Wild • Videos in the wild usually contain multiple scenes • Only a subset of these scenes may contain DeepFakes • Detection process might be slow for multi-shot videos (even short ones) • Low quality videos • Low quality faces tend to fool classifiers • Small detected and fast-moving faces • Usually lead to noisy predictions • Changes in the environment • Moving obstacles in front of the faces • Changes in lighting
  • 17. DeepFake Detection Service @ WeVerify https://www.youtube.com/watch?v=cVljNV V5VPw&ab_channel=TheFakening
  • 18. More details at TTO 2020 Charitidis, P., Kordopatis-Zilos, G., Papadopoulos, S., & Kompatsiaris, Y. (2020). Investigating the impact of preprocessing and prediction aggregation on the DeepFake detection task. Proceedings of the Conference for Truth and Trust Online (TTO) [to appear], https://arxiv.org/abs/2006.07084 https://truthandtrustonline.com/
  • 19. Thank you! Dr. Symeon Papadopoulos papadop@iti.gr @sympap Media Verification (MeVer) https://mever.iti.gr/ @meverteam https://ai4media.eu/ https://weverify.eu/