Open-Set Recognition in the Age of Vision-Language Models

Are vision-language models (VLMs) open-set models because they are trained on internet-scale datasets? We answer this question with a clear no - VLMs introduce closed-set assumptions via their finite query set, making them vulnerable to open-set conditions. We systematically evaluate VLMs for open-set recognition and find they frequently misclassify objects not contained in their query set, leading to alarmingly low precision when tuned for high recall and vice versa. We show that naively increasing the size of the query set to contain more and more classes does not mitigate this problem, but instead causes diminishing task performance and open-set performance. We establish a revised definition of the open-set problem for the age of VLMs, define a new benchmark and evaluation protocol to facilitate standardised evaluation and research in this important area, and evaluate promising baseline approaches based on predictive uncertainty and dedicated negative embeddings on a range of VLM classifiers and object detectors.

Addressing the Challenges of Open-World Object Detection

We address the challenging problem of open world object detection (OWOD), where object detectors must identify objects from known classes while also identifying and continually learning to detect novel objects. Prior work has resulted in detectors that have a relatively low ability to detect novel objects, and a high likelihood of classifying a novel object as one of the known classes. We approach the problem by identifying the three main challenges that OWOD presents and introduce OW-RCNN, an open world object detector that addresses each of these three challenges. OW-RCNN establishes a new state of the art using the open-world evaluation protocol on MS-COCO, showing a drastically increased ability to detect novel objects (16-21% absolute increase in U-Recall), to avoid their misclassification as one of the known classes (up to 52% reduction in A-OSE), and to incrementally learn to detect them while maintaining performance on previously known classes (1-6% absolute increase in mAP).

Uncertainty-Aware Lidar Place Recognition in Novel Environments

State-of-the-art approaches to lidar place recognition degrade significantly when tested on novel environments that are not present in their training dataset. To improve their reliability, we propose uncertainty-aware lidar place recognition, where each predicted place match must have an associated uncertainty that can be used to identify and reject potentially incorrect matches. We introduce a novel evaluation protocol designed to benchmark uncertainty-aware lidar place recognition, and present Deep Ensembles as the first uncertainty-aware approach for this task. Testing across three large-scale datasets and three state-of-the-art architectures, we show that Deep Ensembles consistently improves the performance of lidar place recognition in novel environments. Compared to a standard network, our results show that Deep Ensembles improves the Recall@1 by more than 5% and AuPR by more than 3% on average when tested on previously unseen environments. Our code repository will be made publicly available upon paper acceptance at https://github.com/csiro-robotics/Uncertainty-LPR.

Density-aware NeRF Ensembles: Quantifying Predictive Uncertainty in Neural Radiance Fields

We show that ensembling effectively quantifies model uncertainty in Neural Radiance Fields (NeRFs) if a density-aware epistemic uncertainty term is considered. The naive ensembles investigated in prior work simply average rendered RGB images to quantify the model uncertainty caused by conflicting explanations of the observed scene. In contrast, we additionally consider the termination probabilities along individual rays to identify epistemic model uncertainty due to a lack of knowledge about the parts of a scene unobserved during training. We achieve new state-of-the-art performance across established uncertainty quantification benchmarks for NeRFs, outperforming methods that require complex changes to the NeRF architecture and training regime. We furthermore demonstrate that NeRF uncertainty can be utilised for next-best view selection and model refinement.

Never mind the metrics -- what about the uncertainty? Visualising confusion matrix metric distributions

There are strong incentives to build models that demonstrate outstanding predictive performance on various datasets and benchmarks. We believe these incentives risk a narrow focus on models and on the performance metrics used to evaluate and compare them -- resulting in a growing body of literature to evaluate and compare metrics. This paper strives for a more balanced perspective on classifier performance metrics by highlighting their distributions under different models of uncertainty and showing how this uncertainty can easily eclipse differences in the empirical performance of classifiers. We begin by emphasising the fundamentally discrete nature of empirical confusion matrices and show how binary matrices can be meaningfully represented in a three dimensional compositional lattice, whose cross-sections form the basis of the space of receiver operating characteristic (ROC) curves. We develop equations, animations and interactive visualisations of the contours of performance metrics within (and beyond) this ROC space, showing how some are affected by class imbalance. We provide interactive visualisations that show the discrete posterior predictive probability mass functions of true and false positive rates in ROC space, and how these relate to uncertainty in performance metrics such as Balanced Accuracy (BA) and the Matthews Correlation Coefficient (MCC). Our hope is that these insights and visualisations will raise greater awareness of the substantial uncertainty in performance metric estimates that can arise when classifiers are evaluated on empirical datasets and benchmarks, and that classification model performance claims should be tempered by this understanding.

What's in the Black Box? The False Negative Mechanisms Inside Object Detectors

In object detection, false negatives arise when a detector fails to detect a target object. To understand why object detectors produce false negatives, we identify five 'false negative mechanisms', where each mechanism describes how a specific component inside the detector architecture failed. Focusing on two-stage and one-stage anchor-box object detector architectures, we introduce a framework for quantifying these false negative mechanisms. Using this framework, we investigate why Faster R-CNN and RetinaNet fail to detect objects in benchmark vision datasets and robotics datasets. We show that a detector's false negative mechanisms differ significantly between computer vision benchmark datasets and robotics deployment scenarios. This has implications for the translation of object detectors developed for benchmark datasets to robotics applications.

Uncertainty for Identifying Open-Set Errors in Visual Object Detection

Deployed into an open world, object detectors are prone to a type of false positive detection termed open-set errors. We propose GMM-Det, a real-time method for extracting epistemic uncertainty from object detectors to identify and reject open-set errors. GMM-Det trains the detector to produce a structured logit space that is modelled with class-specific Gaussian Mixture Models. At test time, open-set errors are identified by their low log-probability under all Gaussian Mixture Models. We test two common detector architectures, Faster R-CNN and RetinaNet, across three varied datasets spanning robotics and computer vision. Our results show that GMM-Det consistently outperforms existing uncertainty techniques for identifying and rejecting open-set detections, especially at the low-error-rate operating point required for safety-critical applications. GMM-Det maintains object detection performance, and introduces only minimal computational overhead. We also introduce a methodology for converting existing object detection datasets into specific open-set datasets to consistently evaluate open-set performance in object detection. Code for GMM-Det and the dataset methodology will be made publicly available.

Class Anchor Clustering: a Distance-based Loss for Training Open Set Classifiers

Existing open set classifiers distinguish between known and unknown inputs by measuring distance in a network's logit space, assuming that known inputs cluster closer to the training data than unknown inputs. However, this approach is typically applied post-hoc to networks trained with cross-entropy loss, which neither guarantees nor encourages the hoped-for clustering behaviour. To overcome this limitation, we introduce Class Anchor Clustering (CAC) loss. CAC is an entirely distance-based loss that explicitly encourages training data to form tight clusters around class-dependent anchor points in the logit space. We show that an open set classifier trained with CAC loss outperforms all state-of-the-art techniques on the challenging TinyImageNet dataset, achieving a 2.4% performance increase in AUROC. In addition, our approach outperforms other state-of-the-art distance-based approaches on a number of further relevant datasets. We will make the code for CAC publicly available.

Evaluating Merging Strategies for Sampling-based Uncertainty Techniques in Object Detection

There has been a recent emergence of sampling-based techniques for estimating epistemic uncertainty in deep neural networks. While these methods can be applied to classification or semantic segmentation tasks by simply averaging samples, this is not the case for object detection, where detection sample bounding boxes must be accurately associated and merged. A weak merging strategy can significantly degrade the performance of the detector and yield an unreliable uncertainty measure. This paper provides the first in-depth investigation of the effect of different association and merging strategies. We compare different combinations of three spatial and two semantic affinity measures with four clustering methods for MC Dropout with a Single Shot Multi-Box Detector. Our results show that the correct choice of affinity-clustering combinations can greatly improve the effectiveness of the classification and spatial uncertainty estimation and the resulting object detection performance. We base our evaluation on a new mix of datasets that emulate near open-set conditions (semantically similar unknown classes), distant open-set conditions (semantically dissimilar unknown classes) and the common closed-set conditions (only known classes).

Dropout Sampling for Robust Object Detection in Open-Set Conditions

Dropout Variational Inference, or Dropout Sampling, has been recently proposed as an approximation technique for Bayesian Deep Learning and evaluated for image classification and regression tasks. This paper investigates the utility of Dropout Sampling for object detection for the first time. We demonstrate how label uncertainty can be extracted from a state-of-the-art object detection system via Dropout Sampling. We evaluate this approach on a large synthetic dataset of 30,000 images, and a real-world dataset captured by a mobile robot in a versatile campus environment. We show that this uncertainty can be utilized to increase object detection performance under the open-set conditions that are typically encountered in robotic vision. A Dropout Sampling network is shown to achieve a 12.3% increase in recall (for the same precision score as a standard network) and a 15.1% increase in precision (for the same recall score as the standard network).