My Projects
You can view all my projects in my Github page. Any contributions are welcome!
• Distributed Deep Learning
➤ An End to End Distributed Deep Learning Engine, works both with Streaming and Batch Data built using Apache Flink, TensorFlow, Druid and Kafka.
➤ The whole pipeline is built in 7 very High-Level Steps. Please take a look at this repo for more details.
➤ The above image contains the architecture of the whole pipeline and how it is getting processed along with all technologies used for on-the-fly predictions.
• Nail Tracker
➤ Transfer Learned the feature layers of ResNet Model on Nail Images collected online and deployed this trained model on Android using TensorFlow-Lite for on-the-fly predictions on the camera input.
➤ The GIF clearly shows how well the model is able to predict nails and you can click here to access the Repo.
• Digimon Generator
➤ Developed a custom 8-layered Discriminator and 6-layered Generator Wasserstein Generative Adversarial Network(WGAN) trained on 1072 Digimon images.
➤ This model upon passing any random noise in-turn produces a Unknown Digimon Image which is similar to what the model was trained on but not the same.
➤ Above picture shows how the layers are arranged and click here to access the Repo.
• Image Enhancer
➤ Designed a custom Conditional Generative Adversarial Network(CGAN) trained over 13k Human Face images.
➤ The images were first blurred using Image Augmentation techniques such as Guassian Blur and/or Salt & Pepper noise. Used these images for training the model to predict the actual de-noised images.
➤ Left image is the Blurred image and the Right one is the model predicted de-noised output. Click here to access the Repo.
• Artistic Video Generator
➤ This project takes the Basic Style Transfer to next-level by manipulating the feature layers as well as the addition of some custom CNN layers at the top and end of VGG-16 model to provide the artistic form of an image in fraction of seconds.
➤ Model is developed using TensorFlow and deployed to Android using TF-Lite which receives the input frames from the camera and outputs the predictions (i.e. the artistic forms) on-the-fly and shows the result to the user.
➤ Above GIF shows you how the video is being processed to different artistic forms in an instant while viewing through camera on-the-fly.
• Cross IT - The Game
➤ This is a computer game, which involves the development of various 3D models including the Main Character in Blender and Shaders using ShaderLab & OpenGL.
➤ All these pieces are connected using C# in MonoDevelop to bring the character alive in Unity3D.
➤ The above image is a glimpse of Level 1 where the character has to collect some glowing balls in order to advance to another level. Please click here to access the Repo.