Runze Li

Ph.D candidate

Department of Computer Science and Engineering, UC Riverside.

Email  /  CV  /  Google Scholar

I am a final-year Ph.D in Computer Science and Engineering at the University of California Riverside with Bir Bhanu as my advisor. I obtained my master degree at the University of Melbourne and bachelor degree at the Beijing University of Posts and Telecommunications

I worked as a Research Intern at Google AR, Google Brain (now Google DeepMind), OPPO US Research, UII America, and Siemens.

I'm interested in computer vision, 3D vision, vision-language pre-training, papers are highlighted.

We present RECLIP (Resource-efficient CLIP), a simple method that minimizes computational resource footprint for CLIP (Contrastive Language Image Pretraining). Inspired by the notion of coarse-to-fine in computer vision, we leverage small images to learn from large-scale language supervision efficiently, and finetune the model with high-resolution data in the end.

We propose (1) a depth factorization module to estimate a global depth scale factor, which helps the depth network adapt to the rapid scale changes for indoor environments during model training, (2) a residual pose estimation module that mitigates the inaccurate camera rotation prediction issue in the pose network and in turn improves monocular depth estimation performance, (3) incorporate coordinates convolutional encoding to leverage coordinates cues in inducing relative camera poses.

We propose to (1) visually interpret conditional GANs for facial attribute translation by using a gradient-based attention mechanism. and (2) new learning objectives for knowledge distillation using attention in generative adversarial training, which result in improved synthesized face results, reduced visual confusions and boosted training for GANs in a positive way.

We present (1) a new method for video mesh recovery that divides the human mesh into several local parts following the standard skeletal model, (2) model the dynamics of each local part with separate recurrent models, with each model conditioned appropriately based on the known kinematic structure of the human body and this results in a structure-informed local recurrent learning architecture that can be trained in an end-to-end fashion with available annotations. W

We propose the MonoIndoor, which consists of two novel modules: a depth factorization module and a residual pose estimation module. In the depth factorization module, we factorize the depth map into a global depth scale and a relative depth map. The depth scale factor is separately predicted by an extra branch in the depth network. In the residual pose estimation module, we mitigate the issue of inaccurate rotation prediction by performing residual pose estimation in addition to an initial large pose prediction.

We propose the first technique to visually explain VAEs by means of gradient-based attention. We present methods to generate visual attention from the learned latent space, and also demonstrate such attention explanations serve more than just explaining VAE predictions. We show how these attention maps can be used to localize anomalies in images, and how they can be infused into model training, helping bootstrap the VAE into learning improved latent space disentanglement.

We (1) collect a dataset of highlight videos of soccer players, including two coarse-grained action types of soccer players and six fine-grained actions of players, and provide annotations for the collected dataset, (2) propose an energy-motion features aggregation network-EMA-Net to fully exploit energy-based representation of soccer players movements in video sequences and explicit motion dynamics of soccer players in videos for soccer players’ fine-grained action analysis.

We target on these highlight actions and movements in soccer games and focus on dribbling skills performed by the top players. We leverage understanding of complex dribbling video clips by representing a video sequence with a single Dribble Energy Image(DEI) that is informative for dribbling styles recognition.

Thanks to this awesome template!