Academic research

[strangecosmos]

A thread to post papers and videos about academic research on autonomous vehicles.

[1705.07115] Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics

 

[strangecosmos]

Wow. This video is crazy. GANs generate photorealistic driving videos.

[1808.06601] Video-to-Video Synthesis

Not sure if this is even useful to self-driving car development— maybe for simulation and offline training?

 

[strangecosmos]

Deep Neural Networks Rival the Representation of Primate IT Cortex for Core Visual Object Recognition

 

[strangecosmos]

New paper on 3D object detection using a monocular camera. Abstract:

“3D object detection from monocular images has proven to be an enormously challenging task, with the performance of leading systems not yet achieving even 10% of that of LiDAR-based counterparts. One explanation for this performance gap is that existing systems are entirely at the mercy of the perspective image-based representation, in which the appearance and scale of objects varies drastically with depth and meaningful distances are difficult to infer. In this work we argue that the ability to reason about the world in 3D is an essential element of the 3D object detection task. To this end, we introduce the orthographic feature transform, which enables us to escape the image domain by map- ping image-based features into an orthographic 3D space. This allows us to reason holistically about the spatial con- figuration of the scene in a domain where scale is consistent and distances between objects are meaningful. We apply this transformation as part of an end-to-end deep learn- ing architecture and achieve state-of-the-art performance on the KITTI 3D object benchmark.”
​

[1811.08188] Orthographic Feature Transform for Monocular 3D Object Detection

Source: Alex Kendall on Twitter

 

[CarlK]

Is it only me that have problems with those videos that speed up everything to make them look more capable then what they actually are? Makes my head hurts.

 

[lunitiks]

Is it only me that have problems with those videos that speed up everything to make them look more capable then what they actually are? Makes my head hurts.

Agreed, like the one in here:

Autopilot

 

[CarlK]

Agreed, like the one in here:

Autopilot

Exactly. Anyone could put out impressive edited "demo" videos. These don't mean anything. No?

 

[lunitiks]

Papers on End-to-End Deep Learning for Autonomous Vehicles:

• Bojarski et al., End-to-End Learning for Self-Driving Cars (2016)

• Xu et al., End-to-End Learning of Driving Models from Large-scale Video Datasets (2016/2017)

• Bojarski et al., Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car (2017)

• Yang et al., Feature Analysis and Selection for Training an End-to-End Autonomous Vehicle Controller Using the Deep Learning Approach (2017)

• Chi et al., Deep Steering: Learning End-to-End Driving Model from Spatial and Temporal Visual Cues (2017)

• Codevilla et al., End-to-End Driving via Conditional Imitation Learning (2017/2018)

• Hecker et al., End-to-End Learning of Driving Models with Surround-View Cameras and Route Planners (2018)

• Mehta et al., Learning End-to-End Autonomous Driving using Guided Auxiliary Supervision (2018)

 

[strangecosmos]

Google AI: A Structured Approach to Unsupervised Depth Learning from Monocular Videos

 

[malcolm]

Google AI: A Structured Approach to Unsupervised Depth Learning from Monocular Videos

Thanks for the link. Very interesting.

Importantly, we are able to recover correctly the depth of a car moving at the same speed as the ego-motion vehicle. This has been challenging previously — in this case, the moving vehicle appears (in a monocular input) as static, exhibiting the same behavior as the static horizon, resulting in an inferred infinite depth. While stereo inputs can solve that ambiguity, our approach is the first one that is able to correctly infer that from a monocular input.

 

[strangecosmos]

Berkeley AI Research: Visual Model-Based Reinforcement Learning as a Path towards Generalist Robots