*** 𝗙𝗥𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻: 𝗔𝗿𝗲 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗠𝗼𝗱𝗲𝗹𝘀 𝗥𝗲𝗮𝗱𝘆 𝗳𝗼𝗿 𝗙𝗮𝗰𝗲 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻? *** How 𝗮𝗰𝗰𝘂𝗿𝗮𝘁𝗲 𝗮𝗿𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 𝗶𝗻 𝗳𝗮𝗰𝗲 𝗿𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻? How can we 𝗺𝗮𝗸𝗲 𝘁𝗵𝗲𝗺 𝗲𝘃𝗲𝗻 𝗯𝗲𝘁𝘁𝗲𝗿? Can they help us get rid of the required large face datasets and 𝗴𝗲𝘁 𝗰𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝘃𝗲 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝘄𝗶𝘁𝗵 𝗮 𝘀𝗺𝗮𝗹𝗹 𝘀𝗲𝘁 𝗼𝗳 𝗱𝗮𝘁𝗮? And how to do that? 𝗪𝗵𝗮𝘁 𝗮𝗯𝗼𝘂𝘁 𝗕𝗶𝗮𝘀? All that, in detailed analyses, proposed solutions, and extensive experiments in our new pre-print “FRoundation: Are Foundation Models Ready for Face Recognition?”: https://lnkd.in/e3p_Scwc Follow up for the soon to be released pre-trained models under: https://lnkd.in/e5qBKEif The paper is authored by my colleagues Tahar Chettaoui and Fadi Boutros. Great effort! Abstract: Foundation models are predominantly trained in an unsupervised or self-supervised manner on highly diverse and large-scale datasets, making them broadly applicable to various downstream tasks. In this work, we investigate for the first time whether such models are suitable for the specific domain of face recognition. We further propose and demonstrate the adaptation of these models for face recognition across different levels of data availability. Extensive experiments are conducted on multiple foundation models and datasets of varying scales for training and fine-tuning, with evaluation on a wide range of benchmarks. Our results indicate that, despite their versatility, pre-trained foundation models underperform in face recognition compared to similar architectures trained specifically for this task. However, fine-tuning foundation models yields promising results, often surpassing models trained from scratch when training data is limited. Even with access to large-scale face recognition training datasets, fine-tuned foundation models perform comparably to models trained from scratch, but with lower training computational costs and without relying on the assumption of extensive data availability. Our analysis also explores bias in face recognition, with slightly higher bias observed in some settings when using foundation models. #computervision #biometrics #foundationmodels #visionfoundationmodels #facerecognition #machinelearning #machinelearning #machinevision Technische Universität Darmstadt Fraunhofer IGD ATHENE-Center
Naser Damer’s Post
More Relevant Posts
-
"Innovative Computer Science Student on a Mission to Transform Ideas into Reality 💼"
🎉 Excited to share that our paper titled "Facial Emotion Detection System through Machine Learning Approach" has been published in the International Research Journal of Modernization in Engineering Technology and Science (IRJMETS), Volume 06, Issue 06, June 2024! 🌐📄 In this paper, we present a Facial Emotion Detection System that leverages machine learning techniques to recognize emotions like happiness, sadness, anger, and more, from facial expressions. By integrating a real-time system using computer vision and deep learning, our goal is to enhance human-computer interaction by enabling devices to better understand and respond to human emotions. This work represents months of dedication, collaboration, and passion for using machine learning to enhance human-computer interaction. The journey was filled with learning experiences, and I want to give a special shoutout to my teammates Shubham Shingade and Om Swami couldn't have done this without you! 🙌 A big thank you to our guide Prof. Suraj Dhanawe for your invaluable guidance and support throughout the project. Looking forward to future endeavors and pushing the boundaries of technology! #Research #MachineLearning #FacialEmotionRecognition #Engineering #IRJMETS #Collaboration #Teamwork
-
I wrote a tutorial on diffusion models for undergrad and grad students. I tried my best to give intuitive explanations for complicated equations. Your feedback is much appreciated. Thanks to those who suggested various reading materials to me. https://lnkd.in/gNuzKH3m #machinelearning #generativeAI #computervision
-
This is truly an amazing piece of work and a great read! About 1.5 years ago, when I got into diffusion research, there was a lack of good explanations and resources that clearly broke down concepts for someone very new in the area. This is going to help bridge the understanding gap for so many students and provide a lot of useful insights - thank you so much for this! Some other resources that I found extremely useful for anyone wanting to get started: - Lillian Weng - What are diffusion models? - https://lnkd.in/gNVsuPph - Calvin Luo - Understanding Diffusion Models: A Unified Perspective - https://lnkd.in/gbHgbhzg - Ayan Das - An introduction to Diffusion Probabilistic Models - https://lnkd.in/g34Qw4jT - MIT - Diffusion Probabilistic Models - https://lnkd.in/gD8-YGV2 - Emilio Dorigatti - Simplest Implementation of Diffusion Models - https://lnkd.in/ghNccgVt - Awesome diffusion models (collection of all things diffusion) - https://lnkd.in/g5s9wRtS Some of the stuff mentioned in the collection was very difficult to find or even existent 2 years ago. It's quite insane to witness the progression of research in a field that you're really interested in.
I wrote a tutorial on diffusion models for undergrad and grad students. I tried my best to give intuitive explanations for complicated equations. Your feedback is much appreciated. Thanks to those who suggested various reading materials to me. https://lnkd.in/gNuzKH3m #machinelearning #generativeAI #computervision
Tutorial on Diffusion Models for Imaging and Vision
arxiv.org
-
Probably the most comprehensive tutorial on diffusion models out there right now.
I wrote a tutorial on diffusion models for undergrad and grad students. I tried my best to give intuitive explanations for complicated equations. Your feedback is much appreciated. Thanks to those who suggested various reading materials to me. https://lnkd.in/gNuzKH3m #machinelearning #generativeAI #computervision
Tutorial on Diffusion Models for Imaging and Vision
arxiv.org
-
𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 and 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 systems are rapidly entering numerous aspects of our daily lives, changing the way we produce, process and use information, but also rapidly increasing the need for more computational and electrical power. 𝗡𝗲𝘂𝗿𝗼𝗺𝗼𝗿𝗽𝗵𝗶𝗰 𝗰𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴 is a new approach to create a more efficient, flexible and power-saving computer architecture, by mimicking the human brain. 🧠 The scientists at #FraunhoferEMFT are working on the development of 𝙣𝙚𝙪𝙧𝙤𝙢𝙤𝙧𝙥𝙝𝙞𝙘 𝙘𝙝𝙞𝙥𝙨, which act like neurons of the human brain. One challenge is converting the analog signals from our physical world to a representation which can be understood and processed by such new kind of hardware. This task is addressed by two new Master Theses at Fraunhofer EMFT, now completed and published. 📜 👏🏻 In her Master Thesis “𝘿𝙚𝙨𝙞𝙜𝙣 𝙤𝙛 𝙖𝙣 𝙀𝙫𝙚𝙣𝙩-𝙗𝙖𝙨𝙚𝙙 𝙕𝙚𝙧𝙤-𝘾𝙧𝙤𝙨𝙨𝙞𝙣𝙜 𝘼𝘿𝘾 𝙛𝙤𝙧 𝙄𝙣𝙩𝙚𝙧𝙛𝙖𝙘𝙞𝙣𝙜 𝘼𝙣𝙖𝙡𝙤𝙜 𝙁𝙧𝙤𝙣𝙩𝙚𝙣𝙙 𝙖𝙣𝙙 𝙎𝙉𝙉 𝘼𝙘𝙘𝙚𝙡𝙚𝙧𝙖𝙩𝙤𝙧”, Mariam Vadachkoria explores the design and implementation of an Analog-to-Digital Converter for neuromorphic hardware, converting analog signals into spike-based representations compatible with 𝘀𝗽𝗶𝗸𝗶𝗻𝗴 𝗻𝗲𝘂𝗿𝗮𝗹 𝗻𝗲𝘁𝘄𝗼𝗿𝗸𝘀 (SNN). 🗣📀 Inspired by the cochea in the human ear, Lan Tran analyzes and develops an analog feature extraction frontend in her Master Thesis “𝘿𝙚𝙨𝙞𝙜𝙣 𝙤𝙛 𝙖𝙣 𝘼𝙣𝙖𝙡𝙤𝙜 𝙁𝙚𝙖𝙩𝙪𝙧𝙚 𝙀𝙭𝙩𝙧𝙖𝙘𝙩𝙞𝙤𝙣 𝙛𝙤𝙧 𝙀𝙫𝙚𝙣𝙩-𝙗𝙖𝙨𝙚𝙙 𝙉𝙚𝙪𝙧𝙤𝙢𝙤𝙧𝙥𝙝𝙞𝙘 𝙋𝙧𝙤𝙘𝙚𝙨𝙨𝙤𝙧”, which provides efficient conditioning of audio signals before they are processed by the neuromorphic processor. 🎧 Learn more about the 𝗰𝗶𝗿𝗰𝘂𝗶𝘁 𝗱𝗲𝘀𝗶𝗴𝗻 𝗳𝗼𝗿 𝗻𝗲𝘂𝗿𝗼𝗺𝗼𝗿𝗽𝗵𝗶𝗰 𝗰𝗼𝗺𝗽𝘂𝘁𝗶𝗻𝗴 at Fraunhofer EMFT: https://lnkd.in/dg_bw9aV. You can find the master theses of 𝑴𝒂𝒓𝒊𝒂𝒎 and 𝑳𝒂𝒏 in 𝗙𝗿𝗮𝘂𝗻𝗵𝗼𝗳𝗲𝗿 𝗣𝘂𝗯𝗹𝗶𝗰𝗮: https://lnkd.in/da6kr38w https://lnkd.in/dXCRdFQe #neuromorphiccomputing #circuitdesign #AI #nextgenerationcomputing #masterthesis #SNN #ADC #science #research #womeninscience
-
-
Instructor, Dept. Radiology, Harvard Medical School and Massachusetts General Hospital
Reposting an excellent math tutorial on Variational Autoencoders (VAEs), Denoising Diffusion Probabilistic Models (DDPMs), Score Matching Langevin Dynamics (SMLD), and Stochastic Differential Equations (SDEs)
I wrote a tutorial on diffusion models for undergrad and grad students. I tried my best to give intuitive explanations for complicated equations. Your feedback is much appreciated. Thanks to those who suggested various reading materials to me. https://lnkd.in/gNuzKH3m #machinelearning #generativeAI #computervision
Tutorial on Diffusion Models for Imaging and Vision
arxiv.org
-
I am elated to share the completion of my master's thesis titled 🔎 : "Self-Supervised Pretraining for Hierarchical Image Classification: A Study on Unlabeled Data Utilization." This project represents the culmination of extensive research and practical collaboration with Geberit. I would like to thank Dr. Tilo Sperling for providing this challenging topic as well as constant support and motivation during the tenure of this thesis. I'd also like to emphasize my gratitude to prof Dr. Tom Lahmer, Ms Zouhour Jaouadi Mr. Marc Freibauer, and Ms. Ines Saiger for their invaluable guidance and unwavering support at every critical stage of my thesis. Special thanks are also due to Mr. Alexander Müller from statworx for his expert guidance related to transfer learning principles, which significantly enriched the depth and scope of my work. In a landscape where abundant image data contrasts with scarce labeled data, traditional hierarchical image classification methods often resort to training separate models on limited data at each hierarchy level. This approach not only escalates computational demands but also risks diminished performance. The research proposes a pioneering solution: harnessing a domain-familiar backbone Convolutional Neural Network (CNN), pre-trained on extensive unlabeled image datasets, to generate high-quality embeddings. Leveraging self-supervised learning techniques, notably Momentum Contrast V2 (MoCoV2), we aim to forge a resilient embedding space less susceptible to class distribution imbalances. The results of our experiments showcase significant improvements in precision (0.54 to 0.72), recall (0.50 to 0.73), and F-1 score (0.46 to 0.72) metrics over traditional supervised methods. This underscores the potential of self-supervised learning to revolutionize hierarchical image classification by mitigating data scarcity challenges and enhancing overall performance. This journey has been enriching and rewarding, and I'm excited about the insights gained and the potential impact of this research. I look forward to new opportunities in the future to continue exploring innovative solutions in the field of computer vision and machine learning 🚀 . #MasterThesis #ComputerVision #MachineLearning #Gratitude #ResearchJourney #FutureOpportunities
-
-
📣 New research paper, supported by #Sec4AI4Sec, published in the Journal of Systems and Software 📣 ✒ "Addressing combinatorial experiments and scarcity of subjects by provably orthogonal and crossover experimental designs". 🔍 In Sec4AI4Sec, we deal with some validations with humans to understand how AI methods really help in real life. While doing these experiments with limited numbers of developers/ students, we need a way to design the experiments so that they would still have a significant result. This paper discusses 3 possible balanced designs for this kind of experiment: full factorial design, orthogonal balanced design (taguchi), and crossover balanced design (NEW!). ➡ Find out more: https://lnkd.in/d2Rr6dRd #DesignofExperiments #CrossoverExperimentalDesign #FullFactorialDesign #OrthogonalDesign Fabio Massacci Aurora Papotti Ranindya Paramitha
Paper: Addressing combinatorial experiments and scarcity of subjects by provably orthogonal and crossover experimental designs
https://www.sec4ai4sec-project.eu
-
A new research #paper "Addressing combinatorial experiments and scarcity of subjects by provably orthogonal and crossover experimental designs", supported by Sec4AI4Sec project, was published in the Journal of Systems and Software. This paper, written by Fabio Massacci, Aurora Papotti, Ranindya Paramitha, discusses 3 possible balanced designs for this kind of #experiment: full factorial design, orthogonal balanced design, and crossover balanced design. Find out more: https://lnkd.in/d2Rr6dRd #DesignofExperiments #CrossoverExperimentalDesign #FullFactorialDesign #OrthogonalDesign
📣 New research paper, supported by #Sec4AI4Sec, published in the Journal of Systems and Software 📣 ✒ "Addressing combinatorial experiments and scarcity of subjects by provably orthogonal and crossover experimental designs". 🔍 In Sec4AI4Sec, we deal with some validations with humans to understand how AI methods really help in real life. While doing these experiments with limited numbers of developers/ students, we need a way to design the experiments so that they would still have a significant result. This paper discusses 3 possible balanced designs for this kind of experiment: full factorial design, orthogonal balanced design (taguchi), and crossover balanced design (NEW!). ➡ Find out more: https://lnkd.in/d2Rr6dRd #DesignofExperiments #CrossoverExperimentalDesign #FullFactorialDesign #OrthogonalDesign Fabio Massacci Aurora Papotti Ranindya Paramitha
Paper: Addressing combinatorial experiments and scarcity of subjects by provably orthogonal and crossover experimental designs
https://www.sec4ai4sec-project.eu
-
💥Image classification Hello everyone!! Here I'm introducing my new project Image classification. The shortest way to recognize our own images using one online platform "Teachable Machine".we just need to capture the object through web cam and train it then it will start recognize the images. As an intern at AIMER Society - Artificial Intelligence Medical and Engineering Researchers Society I have been adding a new project to my profile is Teachable Machine #AIMERS #AIMERSOCIETY #APSCHE #AIMERSOCIETY #TEACHABLEMACHINE #MACHINELEARNING #Al #COMPUTRVISION #Giet university #saisatish sir
PhD, Artificial Intelligence Researcher at Facephi, leading Presentation Attacks Detectors. Face, Continuous, and Speaker Authentication Systems.
1moEnrique Mas Candela