Raviteja Vemulapalli
51 Papers
135 Citations
Raviteja Vemulapalli is an academic researcher from Google. The author has contributed to research in topics: Computer science & Convolutional neural network. The author has an hindex of 16, co-authored 36 publications. Previous affiliations of Raviteja Vemulapalli include University of Maryland, College Park & Siemens.
Chat about Author
Papers
•Posted Content
Boosting Image-based Mutual Gaze Detection using Pseudo 3D Gaze
TL;DR: This work proposes a simple and effective approach to boost the performance of image-based mutual gaze detection by using an auxiliary 3D gaze estimation task during training, and achieves the performance boost without additional labeling cost by training the 3 D gaze estimation branch using pseudo 3D gazing labels deduced from mutual gaze labels.
LangDA: Building Context-Awareness via Language for Domain Adaptive Semantic Segmentation
Chang Liu,Bavesh Balaji,Saad Hossain,C. Thomas,Kwei-Herng Lai,Raviteja Vemulapalli,Alexander Wong,Sirisha Rambhatla +7 more
Corpus Synthesis for Zero-shot ASR domain Adaptation using Large Language Models
Hsuan Su,Ting-yao Hu,Hema Swetha Koppula,Raviteja Vemulapalli,Jen-Hao Rick Chang,Karren Yang,G. Mantena,Oncel Tuzel +7 more
- 18 Sep 2023
TL;DR: This paper proposes a novel data synthesis pipeline that uses a Large Language Model (LLM) to generate a target domain text corpus, and a state-of-the-art controllable speech synthesis model to generate the corresponding speech.
Label-efficient Training of Small Task-specific Models by Leveraging Vision Foundation Models
Raviteja Vemulapalli,Hadi Pouransari,Fartash Faghri,Sachin Mehta,Mehrdad Farajtabar,Mohammad Reza Rastegari,Oncel Tuzel +6 more
TL;DR: This work proposes a simple task-oriented knowledge transfer approach that outperforms task-agnostic VFM distillation, ImageNet pretraining and DINO pretraining, and introduces a retrieval-augmented knowledge transfer strategy that uses web-scale image retrieval to curate effective transfer sets.
R3DG features
TL;DR: A new family of 3D skeletal representations for human action recognition, referred to as R3DG features, that explicitly model the 3D geometric relationships between various body parts using rigid body transformations in 3D space and outperforms various state-of-the-art skeleton-based humanaction recognition approaches.