Frédy Houndayi on LinkedIn: #computervision #deeplearning #facialrecognition #opencv #tensorflow… (2024)

🚀 Day 84: Sharing insights on Fine-Tuning Convolutional Neural Networks!Kicking off another exciting day on my #200DaysOfML journey! 🌟 Today, I delved into the power of fine-tuning pre-trained models like VGG16 for transfer learning. Leveraging this technique on the Oxford Flowers 17 dataset has been an enlightening experience!🔍 Insights:👉🏿Why Fine-Tuning?: Adapting ImageNet's learned features to classify diverse flower species with limited data.👉🏿Approach: Cutting the FC head, attaching a new head for our 17 classes, and initially freezing the body.👉🏿Results: Warm-Up Phase at 82.35% accuracy; Final Phase, including unfreezing Conv layers 15+, hitting 95.59% accuracy!🔍 Key Parameters & Environment:Utilizing Python3, scikit-learn 0.19, Keras 2.1, and OpenCV3,TensorFlow🌟 Another step forward in the #200DaysOfML challenge! Fine-tuning CNNs showcases the adaptability of learned features, empowering classification tasks with limited data.GitHub Repo: https://lnkd.in/gTP6ctwe#200DaysOfML #DeepLearning #TransferLearning #NeuralNetworks #MachineLearning #ComputerVision #DataScience #LinkedInLearning

For anyone looking to learn more about creating convolutional neural networks (ConvNets or CNNs) for image classification, I strongly recommend the Computer Vision course on Kaggle (Kaggle.com)!I’m happy to share that I’ve now completed the course and obtained a new certification: Computer Vision (Building Convolutional Neural Networks (ConvNets or CNNs) with TensorFlow and Keras) from Kaggle!#dataanalysis #machinelearning

4

🎉 Completed the Convolutional Neural Networks in TensorFlow course! 🎉In this course, I delved into the fascinating world of CNNs using TensorFlow, a powerful open-source framework for machine learning. I grasped the essential concepts through practical examples behind convolutional neural networks and their pivotal role in computer vision tasks like image classification and object detection.Key Topics Covered:1. Convolutional Layers: I explored how to apply these layers to images, allowing the network to learn intricate features.2. Pooling Techniques: I used pooling to reduce spatial dimensions, enhancing efficiency.3. Model Performance Optimization: I delved into techniques to improve model performance with hyperparameter tuning (including kerastuner, tensorboard).4. Layer Combinations: I experimented with different combinations of convolutional layers.By the course’s end, I had a deep understanding of CNNs and the ability to build and train them effectively using TensorFlow. These newfound skills empower me to create scalable AI-powered algorithms for real-world applications. 🚀#MachineLearning #DeepLearning #TensorFlow #ComputerVision #DeepLearningAI#Coursera#AICommunity #CNN

15

4 Comments

Day 51 of #100DayChallenge, I'm excited to share my latest project: a Cat vs. Dog Classification model using Convolutional Neural Networks (CNNs) in TensorFlow!This project tackles a classic image recognition task using the power of deep learning.Here's a glimpse of the process:Data Preparation:Used TensorFlow Datasets (TFDS) to access a labeled dataset of cat and dog images.Implemented ImageDataGenerator for data augmentation techniques like rotation, shifting, and flipping to improve model robustness.Model Building:Constructed a CNN architecture using Keras with Convolutional and MaxPooling layers to extract features from the images.Added Flatten and Dense layers for classification, incorporating Dropout for regularization to prevent overfitting.Training and Validation:Split the data into training and validation sets to train the model and evaluate its performance.This is a fantastic introduction to computer vision and deep learning concepts.Want to learn more? Let's connect in the comments and discuss the exciting world of image classification!𝐆𝐢𝐭𝐡𝐮𝐛 𝐋𝐢𝐧𝐤 :https://lnkd.in/gRp9KKez#MachineLearning #DeepLearning #ComputerVision #CatsvsDogs #TensorFlow #CNN

7

🚀 Excited to share my latest project! 🚀I've been working on a fun computer vision project that allows users to detect rock, paper, or scissors gestures in real-time using Convolutional Neural Networks (CNN) and computer vision techniques.With the power of deep learning and TensorFlow, the model achieves high accuracy in classifying hand gestures, making it perfect for interactive applications or games.Check out the project on GitHub here : https://lnkd.in/df-wkkj3 and give it a try! Contributions and feedback are always welcome. Let's rock, paper, scissors our way through computer vision! 🎮💻 #ComputerVision #DeepLearning #CNN #RockPaperScissors

249

21 Comments

🐶 I'm excited to share my latest project where I developed an image classification model for dog breed identification using transfer learning with MobileNet. This project involved a series of steps, including preprocessing and augmenting a comprehensive dataset of dog images to enhance the model's performance. By fine-tuning the MobileNet architecture, I achieved high accuracy in identifying various dog breeds. Through this project, I demonstrated my proficiency in image processing, deep learning, and the practical application of advanced neural networks. This work highlights my ability to work with cutting-edge technologies to solve real-world problems. Feel free to check out the details of this project on my GitHub! GitHub link: https://lnkd.in/ga6gpqTfdatasetlink:https://lnkd.in/gDCkVnvB#DeepLearning #MachineLearning #TransferLearning #ImageClassification

18

3 Comments

Here comes my Task 04 Completion 🎉 as #ML_Intern @Prodigy InfoTech! Developed a Hand Gesture Recognition Model using Machine Learning that accurately identifies and classifies different hand gestures from image data 👍🏼 🖐🏼. This model enables intuitive human-computer interaction and gesture-based control systems 🎯 .In this project, i utilized Keras as the deep learning framework and Convolutional Neural Network(CNN) for the design and implementation 📌 .I am delighted to share that i developed a model with full accuracy 🏅 .This accomplishment demonstrates my proficiency in ML technologies.Excited to learn further! 😍 #MachineLearning#DeepLearning#ProdigyInfotech#HandGestureRecognitionGitHub Repository Link: https://lnkd.in/geNdiT9c

12

🎉 Excited to share my latest project! 🎉I've just completed my first GAN (Generative Adversarial Network) model, which generates numbers between 0 to 9 using the MNIST dataset. This project was a fantastic learning experience and a great way to dive deeper into deep learning and generative models.🔍 Project Overview: The GAN model consists of two neural networks, a generator, and a discriminator, which compete against each other to improve their performance. The generator creates images of handwritten digits, and the discriminator evaluates their authenticity. Over time, the generator gets better at producing realistic digits.🛠 Technical Details:Dataset: MNISTFramework: TensorFlow/Keras Epochs: 25000Video: Watch the progression of learning over epochs in the video below!📈 Results: The model successfully learned to generate realistic digits after training for 25000 epochs. Check out the video to see the learning progress and how the generated digits improve over time.💬 I’d love to hear your thoughts and feedback on this project. Feel free to ask any questions or share your own experiences with GANs!You can check github repository at:- https://lnkd.in/dsMEnMj8 #MachineLearning #DeepLearning #AI #ArtificialIntelligence #GAN #MNIST #DataScience #NeuralNetworks #Tech #Programming

37

6 Comments

Latest project on the MNIST dataset using Artificial Neural Networks (ANN) and TensorFlow's Keras! I've been diving deep into machine learning and computer vision, and this project has been an incredible learning experience. The MNIST dataset, a classic in the machine learning community, consists of handwritten digits and is a great starting point for understanding neural networks.In this project, I trained an ANN model using TensorFlow's Keras API to classify handwritten digits with impressive accuracy. The process involved data preprocessing, model architecture design, training, and evaluation. Through meticulous tuning and experimentation, I achieved 97.78 % validation accuracy.Github: https://lnkd.in/gMe7XrV9#MachineLearning #NeuralNetworks #TensorFlow #Keras #DataScience #ComputerVision #MNIST #ArtificialIntelligence #DeepLearning

15

Emerging from the shadows, Hello connections! 🌟 Its been a quite long since I have posted something here After a period of deep learning in the dark, I'm thrilled to share my latest accomplishment that is "Hello World in Neural Network!" 🐾.During this time, I've immersed myself in the world of Convolutional Neural Networks (CNN),Rather than rushing through tutorials, I took the time to master the foundational mathematics behind concepts like 1)What does Neural Network do!? 2)How does it Work!? 3)Gradient Descent Algorithm 4)Backpropagation calculus 5)Mathematics behind Convolution 6)CNN Architecture. 📊💡This is my First Project in Neural Network : “Cat vs Dog Prediction”Using a Kaggle dataset, I developed a robust model featuring 3 CNN layers with expanding neurons, complemented by a 3-layer Artificial Neural Network (ANN) using a sigmoid function for classification. 🧠🌐I'm excited to announce that my model achieved an impressive 93.6% accuracy on both training and test sets, with careful measures in place to prevent overfitting. 🎯Stay tuned for updates on my next project, Your ongoing support means the world to me – thank you! 🙌Dataset Link:https://lnkd.in/gwNv_-wp#deeplearning #codingLane #codebasics#3blue1brown #techgyan #cnn#skct #MachineLearning #NeuralNetworks

14