We designed and trained a dual branch network for Person Re-Identification where residual and attentive branches captures local and long range dependencies respectively.

We designed and trained a CNN network which capture local and global features for multi-resolution person re-identification. We also proposed a resolution based distillation approach to distill resolution related features.

We trained a robust and novel CNN architecture for Vehicle Re-Identification for capturing multiple orientations present in the data. We also do feature distillation for removing background features.

We trained multiple Cycle GAN networks to generate styles between each pair of cameras and then used these generated images to train a siamese network for person re-identification.

We trained a CNN network which is integrated with several attention modules to capture long range dependencies for person re-identification.