Di Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Di Zhang has a total of 95 co-authors. Here are the top co-authors by number of shared publications:
- Pengfei Wan
Kuaishou Technology, Beijing, China
9 shared publications - Chongyang Ma
Kuaishou Technology, Beijing, China
3 shared publications - Haibin Huang
Institute of Artificial Intelligence (TeleAI), China Telecom, Shanghai, China
3 shared publications - Kun Gai
Kuaishou Technology, Beijing, China
3 shared publications - Xintao Wang
Kuaishou Technology
3 shared publications - Chengru Song
Kuaishou Inc., Beijing, China
2 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 14 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Context as Memory: Scene-Consistent Interactive Long Video Generation with Memory Retrieval
- 14 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Top Cited PapersOpen AccessContext as Memory: Scene-Consistent Interactive Long Video Generation with Memory Retrieval
- Open Access
- 14 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
CamCloneMaster: Enabling Reference-based Camera Control for Video Generation
- 14 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Open AccessCamCloneMaster: Enabling Reference-based Camera Control for Video Generation
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 10994-11003
Mavors: Multi-granularity Video Representation for Multimodal Large Language Model
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 10994-11003
Open AccessMavors: Multi-granularity Video Representation for Multimodal Large Language Model
- Open Access
- 3 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 4682-4693
VLM as Policy: Common-Law Content Moderation Framework for Short Video Platform
- 3 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 4682-4693
Open AccessVLM as Policy: Common-Law Content Moderation Framework for Short Video Platform
- Open Access
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Motion Inversion for Video Customization
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessMotion Inversion for Video Customization
- Top Cited PapersOpen Access
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
CineMaster: A 3D-Aware and Controllable Framework for Cinematic Text-to-Video Generation
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Top Cited PapersOpen AccessCineMaster: A 3D-Aware and Controllable Framework for Cinematic Text-to-Video Generation
- Open Access
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-12
GSHeadRelight: Fast Relightability for 3D Gaussian Head Synthesis
- 27 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessGSHeadRelight: Fast Relightability for 3D Gaussian Head Synthesis
- Open Access
- 3 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Towards Unified 3D Hair Reconstruction from Single-View Portraits
- 3 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open AccessTowards Unified 3D Hair Reconstruction from Single-View Portraits
- Open Access
- 20 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 36-51
Kale: Elastic GPU Scheduling for Online DL Model Training
Large-scale GPU clusters have been widely used for effectively training both online and offline deep learning (DL) jobs. However, elastic scheduling in most cases of resource schedulers is dedicated for offline model training where resource adjustment is planned ahead of time. The native autoscaling policy is on the basis of pre-defined threshold and, if applied directly in online model training, often suffers from belated resource adjustment, leading to diminished model accuracy. In this paper, we present Kale, a novel elastic GPU scheduling system to improve the performance of online DL model training. Through traffic forecasting and resource-throughput modeling, Kale automatically pinpoints the number of required GPUs that best accommodate the on-the-fly data samples before performing stabilized autoscaling. An advanced data shuffling strategy is further employed for balancing uneven samples among different training workers, thereby improving the runtime efficacy. Experiments show that Kale substantially outperforms the state-of-the-art solutions. Compared with the default HPA autoscaling strategy, Kale reduces the accumulated lag and downtime by 69.2% and 33.1%, respectively, whilst lowering the SLO violation rate from 19.57% to just 2.6%. Kale has been deployed at Kuaishou's production-level GPU clusters and successfully underpins real-time video recommendation and advertisement at scale.
- 20 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 36-51
Open AccessKale: Elastic GPU Scheduling for Online DL Model Training
Large-scale GPU clusters have been widely used for effectively training both online and offline deep learning (DL) jobs. However, elastic scheduling in most cases of resource schedulers is dedicated for offline model training where resource adjustment is planned ahead of time. The native autoscaling policy is on the basis of pre-defined threshold and, if applied directly in online model training, often suffers from belated resource adjustment, leading to diminished model accuracy. In this paper, we present Kale, a novel elastic GPU scheduling system to improve the performance of online DL model training. Through traffic forecasting and resource-throughput modeling, Kale automatically pinpoints the number of required GPUs that best accommodate the on-the-fly data samples before performing stabilized autoscaling. An advanced data shuffling strategy is further employed for balancing uneven samples among different training workers, thereby improving the runtime efficacy. Experiments show that Kale substantially outperforms the state-of-the-art solutions. Compared with the default HPA autoscaling strategy, Kale reduces the accumulated lag and downtime by 69.2% and 33.1%, respectively, whilst lowering the SLO violation rate from 19.57% to just 2.6%. Kale has been deployed at Kuaishou's production-level GPU clusters and successfully underpins real-time video recommendation and advertisement at scale.
- Open Access
- 28 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 6880-6889
PlacidDreamer: Advancing Harmony in Text-to-3D Generation
Recently, text-to-3D generation has attracted significant attention, resulting in notable performance enhancements. Previous methods utilize end-to-end 3D generation models to initialize 3D Gaussians, multi-view diffusion models to enforce multi-view consistency, and text-to-image diffusion models to refine details with score distillation algorithms. However, these methods exhibit two limitations. Firstly, they encounter conflicts in generation directions since different models aim to produce diverse 3D assets. Secondly, the issue of over-saturation in score distillation has not been thoroughly investigated and solved. To address these limitations, we propose PlacidDreamer, a text-to-3D framework that harmonizes initialization, multi-view generation, and text-conditioned generation with a single multi-view diffusion model, while simultaneously employing a novel score distillation algorithm to achieve balanced saturation. To unify the generation direction, we introduce the Latent-Plane module, a training-friendly plug-in extension that enables multi-view diffusion models to provide fast geometry reconstruction for initialization and enhanced multi-view images to personalize the text-to-image diffusion model. To address the over-saturation problem, we propose to view score distillation as a multi-objective optimization problem and introduce the Balanced Score Distillation algorithm, which offers a Pareto Optimal solution that achieves both rich details and balanced saturation. Extensive experiments validate the outstanding capabilities of our PlacidDreamer. The code is available at https://github.com/HansenHuang0823/PlacidDreamer.
- 28 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 6880-6889
Open AccessPlacidDreamer: Advancing Harmony in Text-to-3D Generation
Recently, text-to-3D generation has attracted significant attention, resulting in notable performance enhancements. Previous methods utilize end-to-end 3D generation models to initialize 3D Gaussians, multi-view diffusion models to enforce multi-view consistency, and text-to-image diffusion models to refine details with score distillation algorithms. However, these methods exhibit two limitations. Firstly, they encounter conflicts in generation directions since different models aim to produce diverse 3D assets. Secondly, the issue of over-saturation in score distillation has not been thoroughly investigated and solved. To address these limitations, we propose PlacidDreamer, a text-to-3D framework that harmonizes initialization, multi-view generation, and text-conditioned generation with a single multi-view diffusion model, while simultaneously employing a novel score distillation algorithm to achieve balanced saturation. To unify the generation direction, we introduce the Latent-Plane module, a training-friendly plug-in extension that enables multi-view diffusion models to provide fast geometry reconstruction for initialization and enhanced multi-view images to personalize the text-to-image diffusion model. To address the over-saturation problem, we propose to view score distillation as a multi-objective optimization problem and introduce the Balanced Score Distillation algorithm, which offers a Pareto Optimal solution that achieves both rich details and balanced saturation. Extensive experiments validate the outstanding capabilities of our PlacidDreamer. The code is available at https://github.com/HansenHuang0823/PlacidDreamer.