A More experimental evaluation In this section, we provide more experimental evaluation on our approach. In Figure 1, we present visualization results of the predicted 3D human mesh from Human3.6M dataset [1]. We can observe that our approach can better handle tiny cues of motion dynamics. Table 1 and Table 2 provide additional ablation results for our approach on the Penn Action dataset [6] when DTW is not applied. Detailed setting of these experiments can be found in Section 4.2 of the main paper.

The increasing number of vision-based tracking systems deployed in production have necessitated fast, robust camera calibration. In the domain of sport, the majority of current work focuses on sports where lines and intersections are easy to extract, and appearance is relatively consistent across venues. However, for more challenging sports like basketball, those techniques are not sufficient. In this paper, we propose an end-to-end approach for single moving camera calibration across challenging scenarios in sports. Our method contains three key modules: 1) area-based court segmentation, 2) camera pose estimation with embedded templates, 3) homography prediction via a spatial transform network (STN). All three modules are connected, enabling end-to-end training. We evaluate our method on a new college basketball dataset and demonstrate state of the art performance in variable and dynamic environments. We also validate our method on the World Cup 2014 dataset to show its competitive performance against the state-of-the-art methods. Lastly, we show that our method is two orders of magnitude faster than the previous state of the art on both datasets.

Given a video of a person in action, we can easily guess the 3D future motion of the person. In this work, we present perhaps the first approach for predicting a future 3D mesh model sequence of a person from past video input. We do this for periodic motions such as walking and also actions like bowling and squatting seen in sports or workout videos. While there has been a surge of future prediction problems in computer vision, most approaches predict 3D future from 3D past or 2D future from 2D past inputs. In this work, we focus on the problem of predicting 3D future motion from past image sequences, which has a plethora of practical applications in autonomous systems that must operate safely around people from visual inputs. Inspired by the success of autoregressive models in language modeling tasks, we learn an intermediate latent space on which we predict the future. This effectively facilitates autoregressive predictions when the input differs from the output domain. Our approach can be trained on video sequences obtained in-the-wild without 3D ground truth labels. The project website with videos can be found at https://jasonyzhang.com/phd.

From an image of a person in action, we can easily guess the 3D motion of the person in the immediate past and future. This is because we have a mental model of 3D human dynamics that we have acquired from observing visual sequences of humans in motion. We present a framework that can similarly learn a representation of 3D dynamics of humans from video via a simple but effective temporal encoding of image features. At test time, from video, the learned temporal representation give rise to smooth 3D mesh predictions. From a single image, our model can recover the current 3D mesh as well as its 3D past and future motion. Our approach is designed so it can learn from videos with 2D pose annotations in a semi-supervised manner. Though annotated data is always limited, there are millions of videos uploaded daily on the Internet. In this work, we harvest this Internet-scale source of unlabeled data by training our model on unlabeled video with pseudo-ground truth 2D pose obtained from an off-the-shelf 2D pose detector. Our experiments show that adding more videos with pseudo-ground truth 2D pose monotonically improves 3D prediction performance. We evaluate our model on the recent challenging dataset of 3D Poses in the Wild and obtain state-of-the-art performance on the 3D prediction task without any fine-tuning. The project website with video can be found at https://akanazawa.github.io/human_dynamics/.

Simultaneously and accurately forecasting the behavior of many interacting agents is imperative for computer vision applications to be widely deployed (e.g., autonomous vehicles, security, surveillance, sports). In this paper, we present a technique using conditional variational autoencoder which learns a model that “personalizes” prediction to individual agent behavior within a group representation. Given the volume of data available and its adversarial nature, we focus on the sport of basketball and show that our approach efficiently predicts context-specific agent motions. We find that our model generates results that are three times as accurate as previous state of the art approaches (5.74 ft vs. 17.95 ft).

A large number of very popular team sports involve the act of one team trying to score a goal against the other. During this game play, defending players constantly try to predict the next move of the attackers to prevent them from scoring, whereas attackers constantly try to predict the next move of the defenders in order to defy them and score. Such behavior is a prime example of the general human faculty to make predictions about the future and is an important facet of human intelligence. An algorithmic solution to learning a model of the external world from sensory inputs in order to make forecasts is an important unsolved problem. In this work we develop a generic framework for forecasting future events in team sports videos directly from visual inputs. We introduce water polo and basketball datasets towards this end and compare the predictions of the proposed methods against expert and non-expert humans.

We propose learning a semantic visual feature representation by training a neural network supervised solely by point and object trajectories in video sequences. Currently, the predominant paradigm for learning visual features involves training deep convolutional networks on an image classification task using very large human-annotated datasets, e.g. ImageNet. Though effective as supervision, semantic image labels are costly to obtain. On the other hand, under high enough frame rates, frame-to-frame associations between the same 3D physical point or an object can be established automatically. By transitivity, such associations grouped into tracks can relate object/point appearance across large changes in pose, illumination and camera viewpoint, providing a rich source of invariance that can be used for training. We train a siamese network we call it AssociationNet to discriminate between correct and wrong associations between patches in different frames of a video sequence. We show that AssociationNet learns useful features when used as pretraining for object recognition in static images, and outperforms random weight initialization and alternative pretraining methods.

We propose the Encoder-Recurrent-Decoder (ERD) model for recognition and prediction of human body pose in videos and motion capture. The ERD model is a recurrent neural network that incorporates nonlinear encoder and decoder networks before and after recurrent layers. We test instantiations of ERD architectures in the tasks of motion capture (mocap) generation, body pose labeling and body pose forecasting in videos. Our model handles mocap training data across multiple subjects and activity domains, and synthesizes novel motions while avoiding drifting for long periods of time. For human pose labeling, ERD outperforms a per frame body part detector by resolving left-right body part confusions. For video pose forecasting, ERD predicts body joint displacements across a temporal horizon of 400ms and outperforms a first order motion model based on optical flow. ERDs extend previous Long Short Term Memory (LSTM) models in the literature to jointly learn representations and their dynamics. Our experiments show such representation learning is crucial for both labeling and prediction in space-time. We find this is a distinguishing feature between the spatio-temporal visual domain in comparison to 1D text, speech or handwriting, where straightforward hard coded representations have shown excellent results when directly combined with recurrent units [31].

We segment moving objects in videos by ranking spatio-temporal segment proposals according to "moving objectness": how likely they are to contain a moving object. In each video frame, we compute segment proposals using multiple figure-ground segmentations on per frame motion boundaries. We rank them with a Moving Objectness Detector trained on image and motion fields to detect moving objects and discard over/under segmentations or background parts of the scene. We extend the top ranked segments into spatio-temporal tubes using random walkers on motion affinities of dense point trajectories. Our final tube ranking consistently outperforms previous segmentation methods in the two largest video segmentation benchmarks currently available, for any number of proposals. Further, our per frame moving object proposals increase the detection rate up to 7% over previous state-of-the-art static proposal methods.

We present a method that segments moving objects in monocular uncalibrated videos using a combination of moving segment proposal generation and moving objectness ranking. We compute segment proposals in each frame by multiple figure-ground segmentations on optical flow boundaries, we call them Moving Object Proposals (MOPs). MOPs increase the object detection rate by 7% over static segment proposals of [23]. MOPs are ranked by a Moving Objectness Detector (MOD) trained from image and motion fields using a multilayer convolutional network. The MOD discards over/under segmentations and focuses on the moving parts of the scene, but remains agnostic to the object categories contained in the video. Finally, the filtered 2D MOPs are extended into temporally coherent spatio-temporal tubes by label diffusion in a dense point trajectory embedding. Our results on the popular Moseg and VSB100 video benchmarks show the proposed method achieves much higher overlap with ground-truth objects using fewer tube proposals than previous works; further, it reaches levels of ground-truth coverage much beyond the maximum possible with existing supervoxel or trajectory clustering approaches.