Computer Vision / Video Analytics

NVIDIA Academic Partners and Inception Members Present AI Research at CVPR 2020

CVPR is one of the main conferences which provide researchers and engineers with the opportunity to meet and discuss their amazing work. This year, with CVPR and other conferences going virtual, we take the opportunity to recognize our academic and Inception industry partners’ work at CVPR 2020 through this post.

Here’s one paper from UC Berkeley researchers: While machine learning at the edge is not a very common trend at CVPR, we could imagine many ways in which edge computing can take on an increasingly important role in this new digital world. One such way is through the creation of more immersive experiences through AR and VR which require heavy on-device processing. If you’re looking for ways to extract every bit of performance out of your GPU for on-device or edge computing, we highly recommend this talk and paper by BAIR, UC Berkeley. 

The authors present a neural network quantization algorithm which heavily reduces memory and speed requirements for computer vision tasks, while leveraging mixed precision and multiple GPUs.

Take a look at these inspiring research projects presented by NVIDIA Academic partners at CVPR 2020.

UC Berkeley (BAIR)Lighthouse: Predicting Lighting Volumes for Spatially-Coherent Illumination
UC Berkeley (BAIR)Stegastamp: Invisible Hyperlinks in Physical Photographs
UC Berkeley (BAIR)CNN-generated images are surprisingly easy to spot…for now
UC Berkeley (BAIR)ZeroQ: A Novel Zero Shot Quantization Framework
UC Berkeley (BAIR)Learning Saliency Propagation for Semi-Supervised Instance Segmentation
UC Berkeley (BAIR)BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning
UC Berkeley (BAIR)Iterative Answer Prediction with Pointer-Augmented Multimodal Transformers for TextVQA
UC Berkeley (BAIR)Something-Else: Compositional Action Recognition with Spatial-Temporal Interaction Networks
UC Berkeley (BAIR)Advisable Learning for Self-driving Vehicles by Internalizing Observation-to-Action Rules
Tsinghua University (Tsinghua SAIL Group)Benchmarking Adversarial Robustness
Peking University (Multimedia Learning Group)Correlating Edge, Pose with Parsing
Peking University (Multimedia Learning Group)Unsupervised Person Re-identification via Multi-label Classification
Peking University (Multimedia Learning Group)Robust Partial Matching for Person Search in the Wild
Peking University (Multimedia Learning Group)AD-Cluster: Augmented Discriminative Clustering for Domain Adaptive Person Re-identification
Max Planck Institute Intelligent Systems (Perceiving Systems Department)VIBE
Max Planck Institute Intelligent Systems (Perceiving Systems Department)Differentiable Volumetric Rendering: Learning Implicit 3D Representations without 3D Supervision
University of Tokyo (Harada Lab)Learning to Optimize Non-Rigid Tracking
University of Tokyo (Harada Lab)Noise Robust Generative Adversarial Networks
Trinity College Dublin (V-SENSE)CatNet: Class Incremental 3D ConvNets for Lifelong Egocentric Gesture Recognition

In addition to the papers featured above, several professors are giving invited talks at one or more of the CVPR workshops. For example, Prof. Andreas Geiger (MPI-IS) is giving several invited talks on his research including his work on 3D representation learning at the 3D Scene Understanding for Vision, Graphics, and Robotics workshop.  To learn more about Prof. Geiger’s work, you may visit his website.

In the AI startup ecosystem,  NVIDIA Inception Premier member Malong technologies has had two papers accepted to this year’s conference. This includes Cross Batch Memory for Embedding Learning (accepted as an oral), and Deformable Siamese Attention Networks for Visual Object Tracking

Malong has also recently released ThermalNet – an AI powered hazard screen system for COVID-19 leveraging NVIDIA Jetson TX2 and Clara Guardian to identify social distancing and elevated body temperatures. 

For more about the CVPR 2020 papers, talks, workshops, and featured projects that NVIDIA research is participating in, visit our CVPR event site.

Discuss (0)