Recognizing the actions performed by humans from their skeletons.
Joint pedestrian detection and attribute recognition with fields and Multi-Task Learning.
PifPaf: Composite Fields for Human Pose estimation, CVPR’19
Adapting fields for detection from aerial images
We have been exploring how to detect humans in the 3D space only using cameras, which are cheap, reliable and ubiquitous. Our major applications are autonomous vehicles and delivery robots. We focused on challenging cases (the long tail) and uncertainty estimation to improve the reliability of autonomous systems.
Revisiting Multi-Task Learning with ROCK: a Deep Residual Auxiliary Block for Visual Detection, NeurIPS’18
Deep Visual Re-Identification with Confidence, 2020
Perceptual Losses for Real-time Style Transfer and Single Image Super-Resolution, ECCV’16
Open Source Library for Human Trajectory prediction (e.g., official code for Social LSTM, Social Gan, etc…)
We introduce a socially-attended attack to assess the social understanding of prediction models.
Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks, CVPR’18
A vehicle trajectory prediction model which leverages both knowledge and data.
Predicting whether pedestrians will stop walking (Stop) or start to walk (Go) in the near future, for better trajectory prediction around road traffic.
A libary for predicting 2D and 3D bounding boxes of humans in autonomous driving scenarios
We build on successful cGAN models to propose a new semantically-aware discriminator that better guides the generator. We aim to learn a shared latent representation that encodes enough information to jointly do semantic segmentation, content reconstruction, along with a coarse-to-fine grained adversarial reasoning.
We propose a collaborative sampling scheme between the generator and discriminator for improved data generation. Guided by the discriminator, our approach refines generated samples through gradient-based optimization in the data (or feature / latent) space, shifting the generator distribution closer to the real data distribution.
DeepSVG: A Hierarchical Generative Network for Vector Graphics Animation