KITTI-360
With the recently released dataset KITTI-360, we develop a set of novel benchmarks to facilitate research at the intersection of vision, graphics and robotics.
Posts by Year
2022
2021
On the Frequency Bias of Generative Models
Does your GAN suffer from a systematic frequency bias?
Animate Implicit Shapes with Forward Skinning
We propose a differentiable forward skinning module to generate implicit shapes in unseen poses.
Novel approaches for fast rendering of neural coordinate-based representations
A quick look at recent progress at using neural coordinate-based representations for real-time applications.
Neural Parts: Learning Expressive 3D Shape Abstractions with Invertible Neural Networks
Abstracting 3D shapes automatically into semantically meaningful parts without any part-level supervision is hard. In this work, we attempt to combine the si...
SMD-Nets: Stereo Mixture Density Networks
We propose a novel stereo matching framework aimed at improving depth accuracy near object boundaries and suited for disparity super-resolution
Sensor Fusion for Self Driving
We use attention-based feature fusion to combine image and LiDAR representations.
How to evaluate tracking with the HOTA metrics
A new family of metrics for evaluating Multi-Object Tracking.
Counterfactual Generative Networks
A generative model structured into independent causal mechanisms produces images for training invariant classifiers.
2020
Generative Radiance Fields for 3D-Aware Image Synthesis
Generative Radiance Fields generate 3D-consistent images, scale well to high resolution and require only unposed 2D images for training.
Label Efficient Visual Abstractions for Autonomous Driving
We analyze the trade-off between annotation time & driving policy performance for several intermediate scene representations.
Convolutional Occupancy Networks
A flexible implicit neural representation to perform large-scale 3D reconstruction.
3D Controllable GANs
We define the new task of 3D controllable image synthesis and propose an approach for solving it by reasoning both in 3D space and in the 2D image domain.
Unsupervised Hierarchical Part-based Decomposition
Within the first year of their life, humans develop a common-sense understanding of the physical behavior of the world. This understanding relies heavily on ...
Robust Driving Across Diverse Weather Conditions in Urban Environments
We develop a novel algorithm to train self-driving vehicles that are able to drive well across a diverse range of weather conditions in urban environments.
Differentiable Volumetric Rendering
Deep neural networks have revolutionized computer vision over the last decade. They excel in 2D-based vision tasks such as object detection, optical flow pre...
Single Image Motion Deblurring
We take advantage of the physical image formation process for self-supervised motion deblurring.
2019
The Road to Safe Self-driving
A simple color patch could severely affect the optical flow prediction systems in autonomous cars.
Texture Fields
Recently, deep learning methods in the 3D domain have gained popularity in the research community. One of the major goals in this area is to reconstruct 3D c...
Occupancy Flow
An intelligent agent that can interact with the world has to be able to reason in 3D. In recent year, there has therefore been a lot of interest in learning-...
Learning to Drive in Cities
We propose a novel self-driving technique which addreses urban scenarios and doesn’t rely on detailed maps. This new approach learns high-level representatio...
DeFuSR: Learning Non-volumetric Depth Fusion using Successive Reprojections
We propose to combine the prior work on multi-view geometry and triangulation with the strength of deep neural networks. To this end, we combine a learning-b...
Superquadrics Revisited
Recent advances in deep learning coupled with the abundance of large shape repositories gave rise to various methods that seek to learn the 3D model of an ob...
Occupancy Networks
In recent years, deep learning has led to many breakthroughs in computer vision. Many tasks such as object detection, semantic segmentation, optical flow est...
Blog Launch
The Autonomous Vision Group at the Max Planck Institute for Intelligent System and the University of Tübingen is excited to launch our new research blog whic...