I work with Prof. Larry Davis in the Computer Vision Lab at UMD. Over the past 2-3 years, my major research focus has been on identifying bottlenecks in visual recognition problems like object and action detection. I have also worked on large-scale video retrieval, distributed training of neural networks and some interesting applications like creating smart thumbnails, instance-aware style transfer and using AI for detecting deception in court-room videos.
- June, 2018: Code for SNIPER, R-FCN-3000 and Soft-Sampling is available.
- May, 2018: SNIPER is on ArXiv. We can train with a batch size of 160 with ResNet-101 on a single GPU node and get 47.6 mAP while processing 5 images/second on COCO.
- March, 2018: SNIP got accepted as an ORAL presentation in CVPR 2018.
- March, 2018: R-FCN-3000 will be presented in CVPR 2018.
- Jan, 2018: DARE is covered in Futurism, DailyMail, Motherboard
- Oct, 2017: SNIP won the Best Student Entry in the COCO 2017 challenge at ICCV.
- Oct, 2017: 8 of the top 15 teams in the COCO 2017 detection challenge used Soft-NMS.
- June, 2017: TCN ranked third in the Activity-Net 2017 proposal generation challenge.