Past Events

Qualifying Exam In Defense of 2D Keypoint Annotations for 3D Animal Reconstruction
× Export Event Save iCal Send to Google Calendar Send to Yahoo Calendar Send to Outlook Live Send to Microsoft Outlook Save iCal Preserve formatting in description (only supported in some calendar applications) Close Friday, April 23, 2021, 02:00pm - 03:30pm

Speaker: Anastasios Stathopoulos

Location : Remote via Zoom

Committee: 

Prof. Dimitris Metaxas (advisor)

Prof. Sungjin Ahn

Prof. Karl Statos

Prof. Yipeng Huang 

Event Type: Qualifying Exam

Abstract: In this talk, we address the problem of learning-based 3D reconstruction of animals from a single image. The main focus of recent work lies at eliminating the reliance of relevant approaches on keypoint annotations, using mostly masks for supervision. For some scenarios, e.g., when it is hard to establish correspondences across instances, this is a reasonable goal. However, we argue that 2D keypoints can get us a long way when it comes to 3D reconstruction and it is important to embrace their significance.More importantly, we observe that current methods for 2D keypoint localization have excellent generalization properties. As a result, even when trained on a small initial training set, the 2D keypoint detector can automatically annotate unlabelled images with robust 2D keypoint detections.We utilize these self-annotated images for training, greatly improving the reconstruction performance of our models.

: 

RU Zoom (link and details)
Link: https://zoom.us/j/9267664973?pwd=ejVzOVlQeTFHNkc5Uk53S2ZId2lDZz09
Meeting ID: 926 766 4973
Passcode: 569269