Alex Bewley

Markus Wulfmeier, Michael Bloesch, Nino Vieillard et al. · deepmind

The majority of language model training builds on imitation learning. It covers pretraining, supervised fine-tuning, and affects the starting conditions for reinforcement learning from human feedback (RLHF). The simplicity and scalability of maximum likelihood estimation (MLE) for next token prediction led to its role as predominant paradigm. However, the broader field of imitation learning can more effectively utilize the sequential structure underlying autoregressive generation. We focus on investigating the inverse reinforcement learning (IRL) perspective to imitation, extracting rewards and directly optimizing sequences instead of individual token likelihoods and evaluate its benefits for fine-tuning large language models. We provide a new angle, reformulating inverse soft-Q-learning as a temporal difference regularized extension of MLE. This creates a principled connection between MLE and IRL and allows trading off added complexity with increased performance and diversity of generations in the supervised fine-tuning (SFT) setting. We find clear advantages for IRL-based imitation, in particular for retaining diversity while maximizing task performance, rendering IRL a strong alternative on fixed SFT datasets even without online data generation. Our analysis of IRL-extracted reward functions further indicates benefits for more robust reward functions via tighter integration of supervised and preference-based LLM post-training.

Tim Salzmann, Lewis Chiang, Markus Ryll et al.

Anticipating the motion of all humans in dynamic environments such as homes and offices is critical to enable safe and effective robot navigation. Such spaces remain challenging as humans do not follow strict rules of motion and there are often multiple occluded entry points such as corners and doors that create opportunities for sudden encounters. In this work, we present a Transformer based architecture to predict human future trajectories in human-centric environments from input features including human positions, head orientations, and 3D skeletal keypoints from onboard in-the-wild sensory information. The resulting model captures the inherent uncertainty for future human trajectory prediction and achieves state-of-the-art performance on common prediction benchmarks and a human tracking dataset captured from a mobile robot adapted for the prediction task. Furthermore, we identify new agents with limited historical data as a major contributor to error and demonstrate the complementary nature of 3D skeletal poses in reducing prediction error in such challenging scenarios.

Georg Heigold, Matthias Minderer, Alexey Gritsenko et al.

We present an architecture and a training recipe that adapts pre-trained open-world image models to localization in videos. Understanding the open visual world (without being constrained by fixed label spaces) is crucial for many real-world vision tasks. Contrastive pre-training on large image-text datasets has recently led to significant improvements for image-level tasks. For more structured tasks involving object localization applying pre-trained models is more challenging. This is particularly true for video tasks, where task-specific data is limited. We show successful transfer of open-world models by building on the OWL-ViT open-vocabulary detection model and adapting it to video by adding a transformer decoder. The decoder propagates object representations recurrently through time by using the output tokens for one frame as the object queries for the next. Our model is end-to-end trainable on video data and enjoys improved temporal consistency compared to tracking-by-detection baselines, while retaining the open-world capabilities of the backbone detector. We evaluate our model on the challenging TAO-OW benchmark and demonstrate that open-world capabilities, learned from large-scale image-text pre-training, can be transferred successfully to open-world localization across diverse videos.

David B. D'Ambrosio, Jonathan Abelian, Saminda Abeyruwan et al.

We present a deep-dive into a real-world robotic learning system that, in previous work, was shown to be capable of hundreds of table tennis rallies with a human and has the ability to precisely return the ball to desired targets. This system puts together a highly optimized perception subsystem, a high-speed low-latency robot controller, a simulation paradigm that can prevent damage in the real world and also train policies for zero-shot transfer, and automated real world environment resets that enable autonomous training and evaluation on physical robots. We complement a complete system description, including numerous design decisions that are typically not widely disseminated, with a collection of studies that clarify the importance of mitigating various sources of latency, accounting for training and deployment distribution shifts, robustness of the perception system, sensitivity to policy hyper-parameters, and choice of action space. A video demonstrating the components of the system and details of experimental results can be found at https://youtu.be/uFcnWjB42I0.

Bingyi Cao, Koert Chen, Kevis-Kokitsi Maninis et al.

Recent progress in vision-language pretraining has enabled significant improvements to many downstream computer vision applications, such as classification, retrieval, segmentation and depth prediction. However, a fundamental capability that these models still struggle with is aligning dense patch representations with text embeddings of corresponding concepts. In this work, we investigate this critical issue and propose novel techniques to enhance this capability in foundational vision-language models. First, we reveal that a patch-level distillation procedure significantly boosts dense patch-text alignment -- surprisingly, the patch-text alignment of the distilled student model strongly surpasses that of the teacher model. This observation inspires us to consider modifications to pretraining recipes, leading us to propose iBOT++, an upgrade to the commonly-used iBOT masked image objective, where unmasked tokens also contribute directly to the loss. This dramatically enhances patch-text alignment of pretrained models. Additionally, to improve vision-language pretraining efficiency and effectiveness, we modify the exponential moving average setup in the learning recipe, and introduce a caption sampling strategy to benefit from synthetic captions at different granularities. Combining these components, we develop TIPSv2, a new family of image-text encoder models suitable for a wide range of downstream applications. Through comprehensive experiments on 9 tasks and 20 datasets, we demonstrate strong performance, generally on par with or better than recent vision encoder models. Code and models are released via our project page at https://gdm-tipsv2.github.io/ .

Connor Schenck, Isaac Reid, Mithun George Jacob et al.

We introduce STRING: Separable Translationally Invariant Position Encodings. STRING extends Rotary Position Encodings, a recently proposed and widely used algorithm in large language models, via a unifying theoretical framework. Importantly, STRING still provides exact translation invariance, including token coordinates of arbitrary dimensionality, whilst maintaining a low computational footprint. These properties are especially important in robotics, where efficient 3D token representation is key. We integrate STRING into Vision Transformers with RGB(-D) inputs (color plus optional depth), showing substantial gains, e.g. in open-vocabulary object detection and for robotics controllers. We complement our experiments with a rigorous mathematical analysis, proving the universality of our methods.

Tim Salzmann, Markus Ryll, Alex Bewley et al.

Visual relationship detection aims to identify objects and their relationships in images. Prior methods approach this task by adding separate relationship modules or decoders to existing object detection architectures. This separation increases complexity and hinders end-to-end training, which limits performance. We propose a simple and highly efficient decoder-free architecture for open-vocabulary visual relationship detection. Our model consists of a Transformer-based image encoder that represents objects as tokens and models their relationships implicitly. To extract relationship information, we introduce an attention mechanism that selects object pairs likely to form a relationship. We provide a single-stage recipe to train this model on a mixture of object and relationship detection data. Our approach achieves state-of-the-art relationship detection performance on Visual Genome and on the large-vocabulary GQA benchmark at real-time inference speeds. We provide ablations, real-world qualitative examples, and analyses of zero-shot performance.

Pei Sun, Weiyue Wang, Yuning Chai et al.

The detection of 3D objects from LiDAR data is a critical component in most autonomous driving systems. Safe, high speed driving needs larger detection ranges, which are enabled by new LiDARs. These larger detection ranges require more efficient and accurate detection models. Towards this goal, we propose Range Sparse Net (RSN), a simple, efficient, and accurate 3D object detector in order to tackle real time 3D object detection in this extended detection regime. RSN predicts foreground points from range images and applies sparse convolutions on the selected foreground points to detect objects. The lightweight 2D convolutions on dense range images results in significantly fewer selected foreground points, thus enabling the later sparse convolutions in RSN to efficiently operate. Combining features from the range image further enhance detection accuracy. RSN runs at more than 60 frames per second on a 150m x 150m detection region on Waymo Open Dataset (WOD) while being more accurate than previously published detectors. As of 11/2020, RSN is ranked first in the WOD leaderboard based on the APH/LEVEL 1 metrics for LiDAR-based pedestrian and vehicle detection, while being several times faster than alternatives.

Jiquan Ngiam, Benjamin Caine, Vijay Vasudevan et al.

Predicting the motion of multiple agents is necessary for planning in dynamic environments. This task is challenging for autonomous driving since agents (e.g. vehicles and pedestrians) and their associated behaviors may be diverse and influence one another. Most prior work have focused on predicting independent futures for each agent based on all past motion, and planning against these independent predictions. However, planning against independent predictions can make it challenging to represent the future interaction possibilities between different agents, leading to sub-optimal planning. In this work, we formulate a model for predicting the behavior of all agents jointly, producing consistent futures that account for interactions between agents. Inspired by recent language modeling approaches, we use a masking strategy as the query to our model, enabling one to invoke a single model to predict agent behavior in many ways, such as potentially conditioned on the goal or full future trajectory of the autonomous vehicle or the behavior of other agents in the environment. Our model architecture employs attention to combine features across road elements, agent interactions, and time steps. We evaluate our approach on autonomous driving datasets for both marginal and joint motion prediction, and achieve state of the art performance across two popular datasets. Through combining a scene-centric approach, agent permutation equivariant model, and a sequence masking strategy, we show that our model can unify a variety of motion prediction tasks from joint motion predictions to conditioned prediction.

Jack Valmadre, Alex Bewley, Jonathan Huang et al.

This paper introduces temporally local metrics for Multi-Object Tracking. These metrics are obtained by restricting existing metrics based on track matching to a finite temporal horizon, and provide new insight into the ability of trackers to maintain identity over time. Moreover, the horizon parameter offers a novel, meaningful mechanism by which to define the relative importance of detection and association, a common dilemma in applications where imperfect association is tolerable. It is shown that the historical Average Tracking Accuracy (ATA) metric exhibits superior sensitivity to association, enabling its proposed local variant, ALTA, to capture a wide range of characteristics. In particular, ALTA is better equipped to identify advances in association independent of detection. The paper further presents an error decomposition for ATA that reveals the impact of four distinct error types and is equally applicable to ALTA. The diagnostic capabilities of ALTA are demonstrated on the MOT 2017 and Waymo Open Dataset benchmarks.

Alex Bewley, Pei Sun, Thomas Mensink et al.

This paper presents a novel 3D object detection framework that processes LiDAR data directly on its native representation: range images. Benefiting from the compactness of range images, 2D convolutions can efficiently process dense LiDAR data of a scene. To overcome scale sensitivity in this perspective view, a novel range-conditioned dilation (RCD) layer is proposed to dynamically adjust a continuous dilation rate as a function of the measured range. Furthermore, localized soft range gating combined with a 3D box-refinement stage improves robustness in occluded areas, and produces overall more accurate bounding box predictions. On the public large-scale Waymo Open Dataset, our method sets a new baseline for range-based 3D detection, outperforming multiview and voxel-based methods over all ranges with unparalleled performance at long range detection.

Alex Bewley, Jessica Rigley, Yuxuan Liu et al.

Simulation can be a powerful tool for understanding machine learning systems and designing methods to solve real-world problems. Training and evaluating methods purely in simulation is often "doomed to succeed" at the desired task in a simulated environment, but the resulting models are incapable of operation in the real world. Here we present and evaluate a method for transferring a vision-based lane following driving policy from simulation to operation on a rural road without any real-world labels. Our approach leverages recent advances in image-to-image translation to achieve domain transfer while jointly learning a single-camera control policy from simulation control labels. We assess the driving performance of this method using both open-loop regression metrics, and closed-loop performance operating an autonomous vehicle on rural and urban roads.

Nicolai Wojke, Alex Bewley

Metric learning aims to construct an embedding where two extracted features corresponding to the same identity are likely to be closer than features from different identities. This paper presents a method for learning such a feature space where the cosine similarity is effectively optimized through a simple re-parametrization of the conventional softmax classification regime. At test time, the final classification layer can be stripped from the network to facilitate nearest neighbor queries on unseen individuals using the cosine similarity metric. This approach presents a simple alternative to direct metric learning objectives such as siamese networks that have required sophisticated pair or triplet sampling strategies in the past. The method is evaluated on two large-scale pedestrian re-identification datasets where competitive results are achieved overall. In particular, we achieve better generalization on the test set compared to a network trained with triplet loss.

Corina Gurau, Alex Bewley, Ingmar Posner

We propose an efficient way to output better calibrated uncertainty scores from neural networks. The Distilled Dropout Network (DDN) makes standard (non-Bayesian) neural networks more introspective by adding a new training loss which prevents them from being overconfident. Our method is more efficient than Bayesian neural networks or model ensembles which, despite providing more reliable uncertainty scores, are more cumbersome to train and slower to test. We evaluate DDN on the the task of image classification on the CIFAR-10 dataset and show that our calibration results are competitive even when compared to 100 Monte Carlo samples from a dropout network while they also increase the classification accuracy. We also propose better calibration within the state of the art Faster R-CNN object detection framework and show, using the COCO dataset, that DDN helps train better calibrated object detectors.

Alex Kendall, Jeffrey Hawke, David Janz et al.

We demonstrate the first application of deep reinforcement learning to autonomous driving. From randomly initialised parameters, our model is able to learn a policy for lane following in a handful of training episodes using a single monocular image as input. We provide a general and easy to obtain reward: the distance travelled by the vehicle without the safety driver taking control. We use a continuous, model-free deep reinforcement learning algorithm, with all exploration and optimisation performed on-vehicle. This demonstrates a new framework for autonomous driving which moves away from reliance on defined logical rules, mapping, and direct supervision. We discuss the challenges and opportunities to scale this approach to a broader range of autonomous driving tasks.

Fabian B. Fuchs, Oliver Groth, Adam R. Kosiorek et al.

Visually predicting the stability of block towers is a popular task in the domain of intuitive physics. While previous work focusses on prediction accuracy, a one-dimensional performance measure, we provide a broader analysis of the learned physical understanding of the final model and how the learning process can be guided. To this end, we introduce neural stethoscopes as a general purpose framework for quantifying the degree of importance of specific factors of influence in deep neural networks as well as for actively promoting and suppressing information as appropriate. In doing so, we unify concepts from multitask learning as well as training with auxiliary and adversarial losses. We apply neural stethoscopes to analyse the state-of-the-art neural network for stability prediction. We show that the baseline model is susceptible to being misled by incorrect visual cues. This leads to a performance breakdown to the level of random guessing when training on scenarios where visual cues are inversely correlated with stability. Using stethoscopes to promote meaningful feature extraction increases performance from 51% to 90% prediction accuracy. Conversely, training on an easy dataset where visual cues are positively correlated with stability, the baseline model learns a bias leading to poor performance on a harder dataset. Using an adversarial stethoscope, the network is successfully de-biased, leading to a performance increase from 66% to 88%.

Michael Tanner, Stefan Saftescu, Alex Bewley et al.

Dense reconstructions often contain errors that prior work has so far minimised using high quality sensors and regularising the output. Nevertheless, errors still persist. This paper proposes a machine learning technique to identify errors in three dimensional (3D) meshes. Beyond simply identifying errors, our method quantifies both the magnitude and the direction of depth estimate errors when viewing the scene. This enables us to improve the reconstruction accuracy. We train a suitably deep network architecture with two 3D meshes: a high-quality laser reconstruction, and a lower quality stereo image reconstruction. The network predicts the amount of error in the lower quality reconstruction with respect to the high-quality one, having only view the former through its input. We evaluate our approach by correcting two-dimensional (2D) inverse-depth images extracted from the 3D model, and show that our method improves the quality of these depth reconstructions by up to a relative 10% RMSE.

Markus Wulfmeier, Alex Bewley, Ingmar Posner

Continuous appearance shifts such as changes in weather and lighting conditions can impact the performance of deployed machine learning models. While unsupervised domain adaptation aims to address this challenge, current approaches do not utilise the continuity of the occurring shifts. In particular, many robotics applications exhibit these conditions and thus facilitate the potential to incrementally adapt a learnt model over minor shifts which integrate to massive differences over time. Our work presents an adversarial approach for lifelong, incremental domain adaptation which benefits from unsupervised alignment to a series of intermediate domains which successively diverge from the labelled source domain. We empirically demonstrate that our incremental approach improves handling of large appearance changes, e.g. day to night, on a traversable-path segmentation task compared with a direct, single alignment step approach. Furthermore, by approximating the feature distribution for the source domain with a generative adversarial network, the deployment module can be rendered fully independent of retaining potentially large amounts of the related source training data for only a minor reduction in performance.

Jeffrey Hawke, Alex Bewley, Ingmar Posner

This paper is about enabling robots to improve their perceptual performance through repeated use in their operating environment, creating local expert detectors fitted to the places through which a robot moves. We leverage the concept of 'experiences' in visual perception for robotics, accounting for bias in the data a robot sees by fitting object detector models to a particular place. The key question we seek to answer in this paper is simply: how do we define a place? We build bespoke pedestrian detector models for autonomous driving, highlighting the necessary trade off between generalisation and model capacity as we vary the extent of the place we fit to. We demonstrate a sizeable performance gain over a current state-of-the-art detector when using computationally lightweight bespoke place-fitted detector models.

Adam R. Kosiorek, Alex Bewley, Ingmar Posner

Class-agnostic object tracking is particularly difficult in cluttered environments as target specific discriminative models cannot be learned a priori. Inspired by how the human visual cortex employs spatial attention and separate "where" and "what" processing pathways to actively suppress irrelevant visual features, this work develops a hierarchical attentive recurrent model for single object tracking in videos. The first layer of attention discards the majority of background by selecting a region containing the object of interest, while the subsequent layers tune in on visual features particular to the tracked object. This framework is fully differentiable and can be trained in a purely data driven fashion by gradient methods. To improve training convergence, we augment the loss function with terms for a number of auxiliary tasks relevant for tracking. Evaluation of the proposed model is performed on two datasets: pedestrian tracking on the KTH activity recognition dataset and the more difficult KITTI object tracking dataset.

Nicolai Wojke, Alex Bewley, Dietrich Paulus

Simple Online and Realtime Tracking (SORT) is a pragmatic approach to multiple object tracking with a focus on simple, effective algorithms. In this paper, we integrate appearance information to improve the performance of SORT. Due to this extension we are able to track objects through longer periods of occlusions, effectively reducing the number of identity switches. In spirit of the original framework we place much of the computational complexity into an offline pre-training stage where we learn a deep association metric on a large-scale person re-identification dataset. During online application, we establish measurement-to-track associations using nearest neighbor queries in visual appearance space. Experimental evaluation shows that our extensions reduce the number of identity switches by 45%, achieving overall competitive performance at high frame rates.

Markus Wulfmeier, Alex Bewley, Ingmar Posner

Appearance changes due to weather and seasonal conditions represent a strong impediment to the robust implementation of machine learning systems in outdoor robotics. While supervised learning optimises a model for the training domain, it will deliver degraded performance in application domains that underlie distributional shifts caused by these changes. Traditionally, this problem has been addressed via the collection of labelled data in multiple domains or by imposing priors on the type of shift between both domains. We frame the problem in the context of unsupervised domain adaptation and develop a framework for applying adversarial techniques to adapt popular, state-of-the-art network architectures with the additional objective to align features across domains. Moreover, as adversarial training is notoriously unstable, we first perform an extensive ablation study, adapting many techniques known to stabilise generative adversarial networks, and evaluate on a surrogate classification task with the same appearance change. The distilled insights are applied to the problem of free-space segmentation for motion planning in autonomous driving.

Alex Bewley, Zongyuan Ge, Lionel Ott et al.

This paper explores a pragmatic approach to multiple object tracking where the main focus is to associate objects efficiently for online and realtime applications. To this end, detection quality is identified as a key factor influencing tracking performance, where changing the detector can improve tracking by up to 18.9%. Despite only using a rudimentary combination of familiar techniques such as the Kalman Filter and Hungarian algorithm for the tracking components, this approach achieves an accuracy comparable to state-of-the-art online trackers. Furthermore, due to the simplicity of our tracking method, the tracker updates at a rate of 260 Hz which is over 20x faster than other state-of-the-art trackers.

ZongYuan Ge, Alex Bewley, Christopher McCool et al.

We present a novel deep convolutional neural network (DCNN) system for fine-grained image classification, called a mixture of DCNNs (MixDCNN). The fine-grained image classification problem is characterised by large intra-class variations and small inter-class variations. To overcome these problems our proposed MixDCNN system partitions images into K subsets of similar images and learns an expert DCNN for each subset. The output from each of the K DCNNs is combined to form a single classification decision. In contrast to previous techniques, we provide a formulation to perform joint end-to-end training of the K DCNNs simultaneously. Extensive experiments, on three datasets using two network structures (AlexNet and GoogLeNet), show that the proposed MixDCNN system consistently outperforms other methods. It provides a relative improvement of 12.7% and achieves state-of-the-art results on two datasets.