Research

Code

DG-Net++: PyTorch implementation of joint disentangling and adaptation for cross-domain person re-id

Wetectron: A platform for weakly-supervised object detection

Dance2Music: PyTorch implementation of a generative model that synthesizes dance from music

DG-Net: PyTorch implementation of joint discriminative and generative learning for person re-id

STEP: PyTorch implementation of a fully end-to-end action detector with no need on extra person detection

PAMTRI: PyTorch implementation of multi-task learning for vehicle re-id using randomized synthetic data

AD: Python implementation of a delay metric to measure video object detection delay

PWC-Net: PyTorch and Caffe implementation of accurate and fast optical flow

MoCoGAN: PyTorch implementation for motion and content decomposed video generation

SNV: C++/Matlab implementation of super normal vector for activity recognition

SSCV: Matlab implementation of super sparse coding vector for action recognition

CascadeSVMs: C++ implementation of an algorithm to handle highly imbalanced large-scale data learning

ActionHOG: C++ implementation of an efficient local spatio-temporal feature for activity recognition

DMM-HOG: Matlab implementation of depth motion maps for action and gesture recognition

Data

CityFlow is a city-scale benchmark for multi-target multi-camera vehicle tracking and re-identification.

Dance2Music is a paired music and dance dataset for cross-modal generation from music to dance.

DG-Market is a large-scale synthetic dataset (10 times larger than the original training set) generated by DG-Net on Market-1501.

SynHead is a large-scale synthetic dataset for video-based head pose estimation.

NVGesture is a multi-modal (color, depth, optical flow and stereo-IR) dataset for online detection and classification of dynamic hand gestures.