• Corpus ID: 204852248

Identifying Unknown Instances for Autonomous Driving

@article{Wong2019IdentifyingUI,
  title={Identifying Unknown Instances for Autonomous Driving},
  author={K. Wong and Shenlong Wang and Mengye Ren and Ming Liang and Raquel Urtasun},
  journal={ArXiv},
  year={2019},
  volume={abs/1910.11296}
}
In the past few years, we have seen great progress in perception algorithms, particular through the use of deep learning. However, most existing approaches focus on a few categories of interest, which represent only a small fraction of the potential categories that robots need to handle in the real-world. Thus, identifying objects from unknown classes remains a challenging yet crucial task. In this paper, we develop a novel open-set instance segmentation algorithm for point clouds which can… 

Figures and Tables from this paper

Multimodal Detection of Unknown Objects on Roads for Autonomous Driving
TLDR
A novel pipeline to detect unknown objects is proposed that makes use of lidar and camera data by combining state-of-the art detection models in a sequential manner and evaluates the approach on the Waymo Open Perception Dataset and points out current research gaps in anomaly detection.
Detecting and Identifying Global Visual Novelties in Driving Scenarios
TLDR
This paper proposes an approach to detect global visual novelties and their corresponding type in real driving scenarios based on pixel-wise and perceptual information and uses a generative adversarial network to detect novelies and a support vector machine to identify their categories.
Video Class Agnostic Segmentation with Contrastive Learning for Autonomous Driving
TLDR
This work addresses the video class agnostic segmentation task, which considers unknown objects outside the closed set of known classes in the authors' training data, and proposes a novel auxiliary contrastive loss to learn the segmentation of knownclasses and unknown objects.
Unsupervised Class-Agnostic Instance Segmentation of 3D LiDAR Data for Autonomous Vehicles
TLDR
A novel approach and a benchmark for class-agnostic instance segmentation for long-tailed classes is proposed and a competitive performance compared to state-of-the-art supervised methods is shown.
Towards Open Set 3D Learning: A Benchmark on Object Point Clouds
TLDR
This paper provides the first broad study on Open Set 3D learning with a novel testbed with settings of increasing complexity in terms of category semantic shift, and investigates the related out-of-distribution and Open Set 2D literature to understand if and how their most recent approaches are effective on 3D data.
Domain Adaptation Through Task Distillation
TLDR
This paper uses image recognition datasets to link up a source and target domain to transfer models between them in a task distillation framework and can successfully transfer navigation policies between drastically different simulators.
Video Class Agnostic Segmentation Benchmark for Autonomous Driving
TLDR
This work formalizes the task of video class agnostic segmentation from monocular video sequences in autonomous driving to account for unknown objects and compares it to a model that uses an auxiliary contrastive loss to improve the discrimination between known and unknown objects.
Benchmarking Open-World LiDAR Instance Segmentation
  • Computer Science
  • 2021
TLDR
A thorough evaluation of existing model-based and data-driven LiDAR instance segmentation methods in the open-world setting and proposes an extension of the SemanticKITTI dataset for benchmarking.
Uncertainty Aware Proposal Segmentation for Unknown Object Detection
  • Yimeng Li, J. Kosecka
  • Computer Science
    2022 IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW)
  • 2022
TLDR
Experimental results demonstrate that the proposed method achieves parallel performance to state of the art methods for unknown object detection and can also be used effectively for reducing object detectors’ false positive rate.
NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds
TLDR
This work introduces a method for instance proposal generation for 3D point clouds based on iterative bilateral bilateral filtering with learned kernels, which considers both the deep feature em-beddings of each point, as well as their locations in the 3D space.
...
...

References

SHOWING 1-10 OF 54 REFERENCES
End-to-End Instance Segmentation with Recurrent Attention
  • Mengye Ren, R. Zemel
  • Computer Science
    2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2017
TLDR
An end-to-end recurrent neural network (RNN) architecture with an attention mechanism to model a human-like counting process, and produce detailed instance segmentations is proposed.
Associative Embedding: End-to-End Learning for Joint Detection and Grouping
TLDR
Associative embedding is introduced, a novel method for supervising convolutional neural networks for the task of detection and grouping for multi-person pose estimation and state-of-the-art performance on the MPII and MS-COCO datasets is reported.
Learning to Segment Every Thing
TLDR
A new partially supervised training paradigm is proposed, together with a novel weight transfer function, that enables training instance segmentation models on a large set of categories all of which have box annotations, but only a small fraction ofWhich have mask annotations.
Learning to Segment Object Candidates
TLDR
A new way to generate object proposals is proposed, introducing an approach based on a discriminative convolutional network that obtains substantially higher object recall using fewer proposals and is able to generalize to unseen categories it has not seen during training.
Instance Segmentation by Jointly Optimizing Spatial Embeddings and Clustering Bandwidth
TLDR
This work proposes a new clustering loss function for proposal-free instance segmentation that pulls the spatial embeddings of pixels belonging to the same instance together and jointly learns an instance-specific clustering bandwidth, maximizing the intersection-over-union of the resulting instance mask.
Semantic Instance Segmentation with a Discriminative Loss Function
TLDR
This work proposes an approach of combining an off-the-shelf network with a principled loss function inspired by a metric learning objective that encourages a convolutional network to produce a representation of the image that can easily be clustered into instances with a simple post-processing step.
Meta-Learning for Semi-Supervised Few-Shot Classification
TLDR
This work proposes novel extensions of Prototypical Networks that are augmented with the ability to use unlabeled examples when producing prototypes, and confirms that these models can learn to improve their predictions due to unlabeling examples, much like a semi-supervised algorithm would.
Deep Watershed Transform for Instance Segmentation
  • Min Bai, R. Urtasun
  • Computer Science
    2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2017
TLDR
This paper presents a simple yet powerful end-to-end convolutional neural network that achieves more than double the performance over the state-of-the-art on the challenging Cityscapes Instance Level Segmentation task.
Proposal-Free Network for Instance-Level Object Segmentation
TLDR
A Proposal-Free Network (PFN) is proposed to address the instance-level object segmentation problem, which outputs the numbers of instances of different categories and the pixel-level information on i) the coordinates of the instance bounding box each pixel belongs to, and ii) the confidences ofDifferent categories for each pixel, based on pixel-to-pixel deep convolutional neural network.
Prototypical Networks for Few-shot Learning
TLDR
This work proposes Prototypical Networks for few-shot classification, and provides an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning.
...
...