Single image-based crowd analytics is plagued with several issues such as occlusion, variations in scale and appearance. We develop novel CNN-based methods such as contextual pyramid CNNs, multi-level fusion scheme, uncertainty aware residual learning, etc to address several of these issues.

Check my projects:

Learning to Count in the Crowd from Limited Labeled Data JHU-CROWD++: Large-Scale Crowd Counting Dataset and A Benchmark Method Pushing the Frontiers of Unconstrained Crowd Counting: New Dataset and Benchmark Method Multi-Level Bottom-Top and Top-Bottom Feature Fusion for Crowd Counting Inverse Attention Guided Deep Crowd Counting Network HA-CCN: Hierarchical Attention-Based Crowd Counting Network A Survey of Recent Advances in CNN-based Single Image Crowd Counting and Density Estimation CNN-based Cascaded Multi-task Learning of High-level prior and Density Estimation for Crowd Counting Generating High-quality Crowd Density Maps using Contextual Pyramid CNNs

Weather conditions such as rain and haze corrupt the image quality, resulting in deterioration of downstream computer vision algorithms. We design specialized CNN networks and loss functions that leverage mathematical models of image degradation to tackle such degradations.

Check my projects:

Syn2Real Transfer Learning for Image Deraining using Gaussian Processes Image De-raining using a Conditional Generative Adversarial Network Joint Transmission Map Estimation and Dehazing using Deep Networks Multi-scale Single Image Dehazing using Perceptual Pyramid Deep Network

Deep networks perform poorly when used on samples that have a distributional shift as compared to that of training samples. We develop unsupervised domain adaptation techniques to adapt networks to different domains for several applications like detection, crowd counting, etc.

Check my projects:

Prior-based Domain Adaptive Object Detection for Hazy and Rainy Conditions Domain Adaptation for Automatic OLED Panel Defect Detection using Adaptive Support Vector Data Description

Object Detection is a critical component in computer vision pipelines like face recognition and autonomous navigation algorithms. We focus on several aspects such as tiny object detection, domain adaptive object detection and fusion architectures for 3D object detection.

Check my projects:

Prior-based Domain Adaptive Object Detection for Hazy and Rainy Conditions MVX-Net: Multimodal VoxelNet for 3D Object Detection DAFE-FD: Density Aware Feature Enrichment for Face Detection Pushing the Limits of Unconstrained Face Detection: a Challenge Dataset and Baseline Results

Face synthesis is crucial for several biometric applications such as face forensics, data augmentation for facial recognition algorithms. We design novel loss functions based on generative adversarial networks for translating between faces and sketches/landmarks.

Check my projects:

GP-GAN: Gender Preserving Gan for Synthesizing Faces from Landmarks High-quality Facial Photo-Sketch Synthesis using Multi-Adversarial Networks