Peng Chang

Xiaohua Qi, Renda Li, Long Peng et al.

Recently Data-Free Knowledge Distillation (DFKD) has garnered attention and can transfer knowledge from a teacher neural network to a student neural network without requiring any access to training data. Although diffusion models are adept at synthesizing high-fidelity photorealistic images across various domains, existing methods cannot be easiliy implemented to DFKD. To bridge that gap, this paper proposes a novel approach based on diffusion models, DiffDFKD. Specifically, DiffDFKD involves targeted optimizations in two key areas. Firstly, DiffDFKD utilizes valuable information from teacher models to guide the pre-trained diffusion models' data synthesis, generating datasets that mirror the training data distribution and effectively bridge domain gaps. Secondly, to reduce computational burdens, DiffDFKD introduces Latent CutMix Augmentation, an efficient technique, to enhance the diversity of diffusion model-generated images for DFKD while preserving key attributes for effective knowledge transfer. Extensive experiments validate the efficacy of DiffDFKD, yielding state-of-the-art results exceeding existing DFKD approaches. We release our code at https://github.com/xhqi0109/DiffDFKD.

Chang Peng, Yifei Zhou, Huifeng Xi et al.

Marker-free human pose estimation (HPE) has found increasing applications in various fields. Current HPE suffers from occasional errors in keypoint recognition and random fluctuation in keypoint trajectories when analyzing kinematic human poses. The performance of existing deep learning-based models for HPE refinement is considerably limited by inaccurate training datasets in which the keypoints are manually annotated. This paper proposed a novel method to overcome the difficulty through joint angle-based modeling. The key techniques include: (i) A joint angle-based model of human pose, which is robust to describe kinematic human poses; (ii) Approximating temporal variation of joint angles through high order Fourier series to get reliable "ground truth"; (iii) A bidirectional recurrent network is designed as a post-processing module to refine the estimation of well-established HRNet. Trained with the high-quality dataset constructed using our method, the network demonstrates outstanding performance to correct wrongly recognized joints and smooth their spatiotemporal trajectories. Tests show that joint angle-based refinement (JAR) outperforms the state-of-the-art HPE refinement network in challenging cases like figure skating and breaking.

Renda Li, Xiaohua Qi, Qiang Ling et al.

Audio-driven cospeech video generation typically involves two stages: speech-to-gesture and gesture-to-video. While significant advances have been made in speech-to-gesture generation, synthesizing natural expressions and gestures remains challenging in gesture-to-video systems. In order to improve the generation effect, previous works adopted complex input and training strategies and required a large amount of data sets for pre-training, which brought inconvenience to practical applications. We propose a simple one-stage training method and a temporal inference method based on a diffusion model to synthesize realistic and continuous gesture videos without the need for additional training of temporal modules.The entire model makes use of existing pre-trained weights, and only a few thousand frames of data are needed for each character at a time to complete fine-tuning. Built upon the video generator, we introduce a new audio-to-video pipeline to synthesize co-speech videos, using 2D human skeleton as the intermediate motion representation. Our experiments show that our method outperforms existing GAN-based and diffusion-based methods.

Peng Chang, Taskin Padir

Manipulation of deformable objects is a desired skill in making robots ubiquitous in manufacturing, service, healthcare, and security. Deformable objects are common in our daily lives, e.g., wires, clothes, bed sheets, etc., and are significantly more difficult to model than rigid objects. In this study, we investigate vision-based manipulation of linear flexible objects such as cables. We propose a geometric modeling method that is based on visual feedback to develop a general representation of the linear flexible object that is subject to gravity. The model characterizes the shape of the object by combining the curvatures on two projection planes. In this approach, we achieve tracking of the position and orientation (pose) of a cable-like object, the pose of its tip, and the pose of the selected grasp point on the object, which enables closed-loop manipulation of the object. We demonstrate the feasibility of our approach by completing the Plug Task used in the 2015 DARPA Robotics Challenge Finals, which involves unplugging a power cable from one socket and plugging it into another. Experiments show that we can successfully complete the task autonomously within 30 seconds.

Peng Chang, Taskin Padir

This paper addresses a new strategy called Simulation-to-Real-to-Simulation (Sim2Real2Sim) to bridge the gap between simulation and real-world, and automate a flexible object manipulation task. This strategy consists of three steps: (1) using the rough environment with the estimated models to develop the methods to complete the manipulation task in the simulation; (2) applying the methods from simulation to real-world and comparing their performance; (3) updating the models and methods in simulation based on the differences between the real world and the simulation. The Plug Task from the 2015 DARPA Robotics Challenge Finals is chosen to evaluate our Sim2Real2Sim strategy. A new identification approach for building the model of the linear flexible objects is derived from real-world to simulation. The automation of the DRC plug task in both simulation and real-world proves the success of the Sim2Real2Sim strategy. Numerical experiments are implemented to validate the simulated model.