DEPARTMENT OF COMPUTER SCIENCE
& ENGINEERING (AI&ML)
Under the guidance of
[Link] Prasad
Review no: 0th ( Abstract)
Project Associates
Mohammad Noorullah (20KT1A4229)
Yallati Siva Naga Lakshmi (20KT1A4261)
Savva Gopi Sree Venkata Tarun (21KT5A4203)
07/02/2024 Department of Computer Science & Engineering (AI&ML)
Project Title-1
Yoga Pose AI Trainer
Department of Computer Science & Engineering (AI&ML) 07/02/2024
ABSTRACT
This paper proposes the development of a
Yoga Pose AI Trainer, an AI-powered system
designed to assist individuals in learning and
perfecting yoga poses.
Existing approaches such as manual
instruction, video tutorials, wearable devices,
and pose recognition software are discussed,
highlighting their limitations in providing
personalized, real-time feedback.
Algorithms including Convolutional Neural
Networks (CNNs), Support Vector Machines
(SVMs), and Recurrent Neural Networks
(RNNs) are utilized for pose detection,
classification, and feedback generation.
Department of Computer Science & Engineering (AI&ML) 07/02/2024
Existing Approaches
Wearable Devices
Video Tutorials
Manual Instruction
Pose Recognition Software
Department of Computer Science & Engineering (AI&ML) 03/31/2024
Gaps and Limitations:
Lack of Personalization: Existing
approaches often provide generic feedback
and may not address individual needs or
limitations.
Limited Feedback: Manual instruction and
pre-recorded videos cannot offer real-time
feedback during practice.
Accuracy Issues: Wearable devices and
pose recognition software may suffer from
accuracy issues, especially in complex
poses or with varying lighting conditions.
Department of Computer Science & Engineering (AI&ML) 03/31/2024
Requirements
Hardware List Software List
System/Laptop Python IDLE
8Gb of RAM Jupyter Notebook
Processor (Intel i5) Visual Studio
GPU (1060) Chrome Browser
Department of Computer Science & Engineering (AI&ML) 07/02/2024
Project-2
Deep Face
Department of Computer Science & Engineering (AI&ML) 07/02/2024
ABSTRACT
Facial recognition technology has Our approach integrates advanced This research contributes to the
seen significant advancements with data augmentation techniques, advancement of deep face technology
the advent of deep learning adversarial training, attention while addressing critical societal
techniques. However, existing mechanisms, and federated learning concerns related to privacy, fairness,
approaches are challenged by issues to enhance the robustness, fairness, and bias.
such as data bias, privacy concerns, and privacy of facial recognition
and robustness to variations. In this systems
study, we propose a comprehensive
methodology to address these
challenges.
Department of Computer Science & Engineering (AI&ML) 07/02/2024
Existing Approaches
Transfer Learning
Generative Adversarial Networks (GANs)
Convolutional Neural Networks (CNNs)
Department of Computer Science & Engineering (AI&ML) 03/31/2024
Gaps and Limitations
Data Bias: Many existing facial recognition datasets suffer from
biases in terms of demographics (e.g., race, gender), leading to
biased models that perform poorly on underrepresented groups.
Privacy Concerns: Facial recognition systems raise privacy
concerns due to the potential for misuse and unauthorized
surveillance.
Ethical Concerns: There are ethical concerns surrounding the use
of facial recognition technology, including potential misuse for
surveillance, discrimination, and invasion of privacy.
Department of Computer Science & Engineering (AI&ML) 03/31/2024
Requirements
Hardware List Software List
System/Laptop Python IDLE
8Gb of RAM Jupyter Notebook
Processor (Intel i5) Visual Studio
GPU (1060) Chrome Browser
Department of Computer Science & Engineering (AI&ML) 07/02/2024
Project-3
Skin Cancer detection
Department of Computer Science & Engineering (AI&ML) 07/02/2024
ABSTRACT
Skin cancer is a prevalent and potentially life-threatening disease, necessitating accurate and
timely diagnosis. In recent years, computer vision techniques, particularly deep learning
models, have shown promise in automating skin lesion classification.
This study proposes a novel approach utilizing the You Only Look Once (YOLO) model, a
state-of-the-art object detection algorithm, for skin cancer classification. The proposed
methodology involves transfer learning, where a pre-trained YOLO model is fine-tuned on a
dataset of skin lesion images. By leveraging transfer learning, the model can adapt its learned
features to the task of skin cancer classification, mitigating the need for extensive labeled
data and computational resources.
The system's performance is evaluated on various metrics, including accuracy, precision,
recall, and F1-score, demonstrating its effectiveness in accurately identifying different types
of skin lesions. Additionally, interpretability techniques such as Grad-CAM are employed to
enhance the transparency and trustworthiness of the model's predictions.
Department of Computer Science & Engineering (AI&ML) 07/02/2024
Existing Approaches
Deep Learning Approaches
Traditional Machine Learning Techniques
Department of Computer Science & Engineering (AI&ML) 03/31/2024
Gaps and Limitations
Limited Dataset Availability: One of the primary challenges
in skin cancer classification is the availability of high-
quality labeled datasets, especially for rare types of skin
lesions.
Interpretability: Deep learning models, including YOLO, are
often criticized for their lack of interpretability, making it
challenging to understand the reasoning behind their
predictions, which is crucial in medical applications.
Computational Complexity: YOLO models are
computationally intensive, especially during training, which
can be a limitation in resource-constrained environments.
Department of Computer Science & Engineering (AI&ML) 03/31/2024
Requirements
Hardware List Software List
System/Laptop Python IDLE
8Gb of RAM Jupyter Notebook
Processor (Intel i5) Visual Studio
GPU (1060) Chrome Browser
Department of Computer Science & Engineering (AI&ML) 07/02/2024
Thank
You
Department of Computer Science & Engineering (AI&ML) 03/31/2024
