top of page
EzFlow 7.png
EzFlow
 
A PyTorch library for Optical Flow Estimation.
 
  • The library supports 6 models with more coming in the future. It also provides modular components for Encoder, Cost Volume and Decoder, and a registry system to register these components. This enables multiple configurability options for the research and development of optical flow models.
     
  • Supported Datasets: FlyingChairs, SceneFlow, Sintel, Kitti, HD1K and Kubric.
     
  • Pre-trained checkpoints available for FlowNetC, PWCNet, RAFT and DCVNet.
     

GitHub
A Case Study on Image Inpainting

Implemented a case study to visualize the effectiveness of Random Box Masks versus Free-Form Masks and various Normalization methods in GANs for the Image Inpainting task. 

DeepFillV2 is used as the baseline model. The results of the case study are visually compared and also evaluated with PSNR and SSIM metrics.


GitHub


Image source: Places365 dataset
 
Emotion Detection in a Smartphone
 
A prototype that can detect emotions based on facial expression using Computer Vision and Machine Learning. It uses OpenCV and Support Vector Machines
to classify the emotion states.
 
The training and predictions are done on-device keeping user privacy in mind.
bottom of page