This artifact aims to provide source code, benchmark suite, results, and materials used in our study “Keeper: Automated Testing and Fixing of Machine Learning Software” [ 3 ]. We developed an automated testing and fixing tool Keeper and its IDE plugin for ML software. It automatically detects software defects and attempts to change how ML APIs are used to alleviate software misbehavior. This artifact provides guidelines to set up and execute Keeper and also guidelines to interpret our evaluation results. We hope this artifact can motivate and help future research to further tackle ML API misuses. All related data are available online.

We introduce nested diffusion models, an efficient and powerful hierarchical generative framework that substantially enhances the generation quality of diffusion models, particularly for images of complex scenes. Our approach employs a series of diffusion models to progressively generate latent variables at different semantic levels. Each model in this series is conditioned on the output of the preceding higher-level models, culminating in image generation. Hierarchical latent variables guide the generation process along predefined semantic pathways, allowing our approach to capture intricate structural details. To construct these latent variables, we leverage a pre-trained visual encoder, which learns strong semantic visual representations, and modulate its capacity via dimensionality reduction and noise injection. Across multiple datasets, our system demonstrates significant enhancements in image quality for both unconditional and class/text conditional generation. Moreover, our unconditional generation system substantially outperforms the baseline conditional system. These advancements incur minimal computational overhead as the more abstract levels of our hierarchy work with lower-dimensional representations.

We present PROGRESSOR, a novel framework that learns a task-agnostic reward function from videos, enabling policy training through goal-conditioned reinforcement learning (RL) without manual supervision. Underlying this reward is an estimate of the distribution over task progress as a function of the current, initial, and goal observations that is learned in a self-supervised fashion. Crucially, PROGRESSOR refines rewards adversarially during online RL training by pushing back predictions for out-of-distribution observations, to mitigate distribution shift inherent in non-expert observations. Utilizing this progress prediction as a dense reward together with an adversarial push-back, we show that PROGRESSOR enables robots to learn complex behaviors without any external supervision. Pretrained on large-scale egocentric human video from EPIC-KITCHENS, PROGRESSOR requires no fine-tuning on in-domain task-specific data for generalization to real-robot offline RL under noisy demonstrations, outperforming contemporary methods that provide dense visual reward for robotic learning. Our findings highlight the potential of PROGRESSOR for scalable robotic applications where direct action labels and task-specific rewards are not readily available.

Neural PDE solvers offer a powerful tool for modeling complex dynamical systems, but often struggle with error accumulation over long time horizons and maintaining stability and physical consistency. We introduce a multiscale implicit neural emulator that enhances long-term prediction accuracy by conditioning on a hierarchy of lower-dimensional future state representations. Drawing inspiration from the stability properties of numerical implicit time-stepping methods, our approach leverages predictions several steps ahead in time at increasing compression rates for next-timestep refinements. By actively adjusting the temporal downsampling ratios, our design enables the model to capture dynamics across multiple granularities and enforce long-range temporal coherence. Experiments on turbulent fluid dynamics show that our method achieves high short-term accuracy and produces long-term stable forecasts, significantly outperforming autoregressive baselines while adding minimal computational overhead.

We present an approach for analyzing grouping information contained within a neural network's activations, per-mitting extraction of spatial layout and semantic segmentation from the behavior of large pre-trained vision models. Unlike prior work, our method conducts a who lis tic analysis of a network's activation state, leveraging features from all layers and obviating the need to guess which part of the model contains relevant information. Motivated by classic spectral clustering, we formulate this analysis in terms of an optimization objective involving a set of affinity matrices, each formed by comparing features within a different layer. Solving this optimization problem using gradient descent allows our technique to scale from single images to dataset-level analysis, including, in the latter, both intra-and inter-image relationships. Analyzing a pre-trained generative transformer provides insight into the computational strategy learned by such models. Equating affinity with key-query similarity across attention layers yields eigenvectors encoding scene spatial layout, whereas defining affinity by value vector similarity yields eigenvectors encoding object identity. This result suggests that key and query vectors co-ordinate attentional information flow according to spatial proximity (a ‘where’ pathway), while value vectors refine a semantic category representation (a ‘what’ pathway).

We introduce HyperFields, a method for generating text-conditioned Neural Radiance Fields (NeRFs) with a single forward pass and (optionally) some fine-tuning. Key to our approach are: (i) a dynamic hypernetwork, which learns a smooth mapping from text token embeddings to the space of NeRFs; (ii) NeRF distillation training, which distills scenes encoded in individual NeRFs into one dynamic hypernetwork. These techniques enable a single network to fit over a hundred unique scenes. We further demonstrate that HyperFields learns a more general map between text and NeRFs, and consequently is capable of predicting novel in-distribution and out-of-distribution scenes — either zero-shot or with a few finetuning steps. Finetuning HyperFields benefits from accelerated convergence thanks to the learned general map, and is capable of synthesizing novel scenes 5 to 10 times faster than existing neural optimization-based methods. Our ablation experiments show that both the dynamic architecture and NeRF distillation are critical to the expressivity of HyperFields.

We develop a neural network architecture which, trained in an unsupervised manner as a denoising diffusion model, simultaneously learns to both generate and segment images. Learning is driven entirely by the denoising diffusion objective, without any annotation or prior knowledge about regions during training. A computational bottleneck, built into the neural architecture, encourages the denoising network to partition an input into regions, denoise them in parallel, and combine the results. Our trained model generates both synthetic images and, by simple examination of its internal predicted partitions, semantic segmentations of those images. Without fine-tuning, we directly apply our unsupervised model to the downstream task of segmenting real images via noising and subsequently denoising them. Experiments demonstrate that our model achieves accurate unsupervised image segmentation and high-quality synthetic image generation across multiple datasets.

Image relighting is the task of showing what a scene from a source image would look like if illuminated differently. Inverse graphic schemes recover an explicit representation of geometry and a set of chosen intrinsics, then relight with some form of renderer. But error control for inverse graphics is difficult, and inverse graphics methods can represent only the effects of the chosen intrinsics. This paper describes a relighting method that is entirely data-driven, where intrinsics and lighting are each represented as latent variables. Our approach produces SOTA relightings of real scenes, as measured by standard metrics. We show that albedo can be recovered from our latent intrinsics without using any example albedos, and that the albedos recovered are competitive with SOTA methods.

As large language models (LLMs) take on complex tasks, their inputs are supplemented with longer contexts that incorporate domain knowledge. Yet using long contexts is challenging as nothing can be generated until the whole context is processed by the LLM. While the context-processing delay can be reduced by reusing the KV cache of a context across different inputs, fetching the KV cache, which contains large tensors, over the network can cause high extra network delays.

Resistive random-access memory (RRAM) is widely recognized as a promising emerging hardware platform for deep neural networks (DNNs). Yet, due to manufacturing limitations, current RRAM devices are highly susceptible to hardware defects, which poses a significant challenge to their practical applicability. In this paper, we present a machine learning technique that enables the deployment of defect-prone RRAM accelerators for DNN applications, without necessitating modifying the hardware, retraining of the neural network, or implementing additional detection circuitry/logic. The key idea involves incorporating a drop-connect inspired approach during the training phase of a DNN, where random subsets of weights are selected to emulate fault effects (e.g., set to zero to mimic stuck-at-1 faults), thereby equipping the DNN with the ability to learn and adapt to RRAM defects with the corresponding fault rates. Our results demonstrate the viability of the dropconnect approach, coupled with various algorithm and system-level design and trade-off considerations. We show that, even in the presence of high defect rates (e.g., up to 30%), the degradation of DNN accuracy can be as low as less than 1% compared to that of the fault-free version, while incurring minimal system-level runtime/energy costs.

The increasing number of software applications incorporating machine learning (ML) solutions has led to the need for testing techniques. However, testing ML software requires tremendous human effort to design realistic and relevant test inputs and to judge software output correctness according to human common sense. Even when misbehavior is exposed, it is often unclear whether the defect is inside ML API or the surrounding code and how to fix the implementation. This article tackles these challenges by proposing Keeper, an automated testing and fixing tool for ML software. The core idea of Keeper is designing pseudo-inverse functions that semantically reverse the corresponding ML task in an empirical way and proxy common human judgment of real-world data. It incorporates these functions into a symbolic execution engine to generate tests. Keeper also detects code smells that degrade software performance. Once misbehavior is exposed, Keeper attempts to change how ML APIs are used to alleviate the misbehavior. Our evaluation on a variety of applications shows that Keeper greatly improves branch coverage, while identifying 74 previously unknown failures and 19 code smells from 56 out of 104 applications. Our user studies show that 78% of end-users and 95% of developers agree with Keeper’s detection and fixing results.

We develop an approach to efficiently grow neural networks, within which parameterization and optimization strategies are designed by considering their effects on the training dynamics. Unlike existing growing methods, which follow simple replication heuristics or utilize auxiliary gradient-based local optimization, we craft a parameterization scheme which dynamically stabilizes weight, activation, and gradient scaling as the architecture evolves, and maintains the inference functionality of the network. To address the optimization difficulty resulting from imbalanced training effort distributed to subnetworks fading in at different growth phases, we propose a learning rate adaption mechanism that rebalances the gradient contribution of these separate subcomponents. Experiments show that our method achieves comparable or better accuracy than training large fixed-size models, while saving a substantial portion of the original training computation budget. We demonstrate that these gains translate into real wall-clock training speedups.

Due to the under-specified interfaces, developers face challenges in correctly integrating machine learning (ML) APIs in software. Even when the ML API and the software are well designed on their own, the resulting application misbehaves when the API output is incompatible with the software. It is desirable to have an adapter that converts ML API output at runtime to better fit the software need and prevent integration failures. In this paper, we conduct an empirical study to understand ML API integration problems in real-world applications. Guided by this study, we present SmartGear, a tool that automatically detects and converts mismatching or incorrect ML API output at run time, serving as a middle layer between ML API and software. Our evaluation on a variety of open-source applications shows that SmartGear detects 70% incompatible API outputs and prevents 67% potential integration failures, outperforming alternative solutions.

An increasing number of software applications incorporate machine learning (ML) solutions for cognitive tasks that statistically mimic human behaviors. To test such software, tremendous human effort is needed to design image/text/audio inputs that are relevant to the software, and to judge whether the software is processing these inputs as most human beings do. Even when misbehavior is exposed, it is often unclear whether the culprit is inside the cognitive ML API or the code using the API.

We study the problem of training deep networks while quantizing parameters and activations into low-precision numeric representations, a setting central to reducing energy consumption and inference time of deployed models. We propose a method that learns different precisions, as measured by bits in numeric representations, for different weights in a neural network, yielding a heterogeneous allocation of bits across parameters. Learning precisions occurs alongside learning weight values, using a strategy derived from a novel framework wherein the intractability of optimizing discrete precisions is approximated by training per-parameter noise magnitudes. We broaden this framework to also encompass learning precisions for hidden state activations, simultaneously with weight precisions and values. Our approach exposes the objective of constructing a low-precision inference-efficient model to the entirety of the training process. Experiments show that it finds highly heterogeneous precision assignments for CNNs trained on CIFAR and ImageNet, improving upon previous state-of-the-art quantization methods. Our improvements extend to the challenging scenario of learning reduced-precision GANs.

Self-supervised representation learning has made significant leaps fueled by progress in contrastive learning, which seeks to learn transformations that embed positive input pairs nearby, while pushing negative pairs far apart. While positive pairs can be generated reliably (e.g., as different views of the same image), it is difficult to accurately establish negative pairs, defined as samples from different images regardless of their semantic content or visual features. A fundamental problem in contrastive learning is mitigating the effects of false negatives. Contrasting false negatives induces two critical issues in representation learning: discarding semantic information and slow convergence. In this paper, we propose novel approaches to identify false negatives, as well as two strategies to mitigate their effect, i.e. false negative elimination and attraction, while systematically performing rigorous evaluations to study this problem in detail. Our method exhibits consistent improvements over existing contrastive learning-based methods. Without labels, we identify false negatives with ~40% accuracy among 1000 semantic classes on ImageNet, and achieve 5.8% absolute improvement in top-1 accuracy over the previous state-of-the-art when finetuning with 1% labels. Our code is available at https://github.com/google-research/fnc

We study image segmentation from an information-theoretic perspective, proposing a novel adversarial method that performs unsupervised segmentation by partitioning images into maximally independent sets. More specifically, we group image pixels into foreground and background, with the goal of minimizing predictability of one set from the other. An easily computed loss drives a greedy search process to maximize inpainting error over these partitions. Our method does not involve training deep networks, is computationally cheap, class-agnostic, and even applicable in isolation to a single unlabeled image. Experiments demonstrate that it achieves a new state-of-the-art in unsupervised segmentation quality, while being substantially faster and more general than competing approaches.1

From a simplified analysis of adaptive methods, we derive AvaGrad, a new optimizer which outperforms SGD on vision tasks when its adaptability is properly tuned. We observe that the power of our method is partially explained by a decoupling of learning rate and adaptability, greatly simplifying hyperparameter search. In light of this observation, we demonstrate that, against conventional wisdom, Adam can also outperform SGD on vision tasks, as long as the coupling between its learning rate and adaptability is taken into account. In practice, AvaGrad matches the best results, as measured by generalization accuracy, delivered by any existing optimizer (SGD or adaptive) across image classification (CIFAR, ImageNet) and character-level language modelling (Penn Treebank) tasks. When training GANs, AvaGrad improves upon existing optimizers.1

We develop an approach to learning visual representations that embraces multimodal data, driven by a combination of intra- and inter-modal similarity preservation objectives. Unlike existing visual pre-training methods, which solve a proxy prediction task in a single domain, our method exploits intrinsic data properties within each modality and semantic information from cross-modal correlation simultaneously, hence improving the quality of learned visual representations. By including multimodal training in a unified framework with different types of contrastive losses, our method can learn more powerful and generic visual features. We first train our model on COCO and evaluate the learned visual representations on various downstream tasks including image classification, object detection, and instance segmentation. For example, the visual representations pre-trained on COCO by our method achieve state-of-the-art top-1 validation accuracy of 55.3% on ImageNet classification, under the common transfer protocol. We also evaluate our method on the large-scale Stock images dataset and show its effectiveness on multi-label image tagging, and cross-modal retrieval tasks.

Machine learning (ML) cloud APIs enable developers to easily incorporate learning solutions into software systems. Unfortunately, ML APIs are challenging to use correctly and efficiently, given their unique semantics, data requirements, and accuracy-performance tradeoffs. Much prior work has studied how to develop ML APIs or ML cloud services, but not how open-source applications are using ML APIs. In this paper, we manually studied 360 representative open-source applications that use Google or AWS cloud-based ML APIs, and found 70% of these applications contain API misuses in their latest versions that degrade functional, performance, or economical quality of the software. We have generalized 8 anti-patterns based on our manual study and developed automated checkers that identify hundreds of more applications that contain ML API misuses.

The core of our approach, Pixel Consensus Voting, is a framework for instance segmentation based on the generalized Hough transform. Pixels cast discretized, probabilistic votes for the likely regions that contain instance centroids. At the detected peaks that emerge in the voting heatmap, backprojection is applied to collect pixels and produce instance masks. Unlike a sliding window detector that densely enumerates object proposals, our method detects instances as a result of the consensus among pixel-wise votes. We implement vote aggregation and backprojection using native operators of a convolutional neural network. The discretization of centroid voting reduces the training of instance segmentation to pixel labeling, analogous and complementary to FCN-style semantic segmentation, leading to an efficient and unified architecture that jointly models things and stuff. We demonstrate the effectiveness of our pipeline on COCO and Cityscapes Panoptic Segmentation and obtain competitive results. Code will be open-sourced.

We introduce a novel approach to endowing neural networks with emergent, long-term, large-scale memory. Distinct from strategies that connect neural networks to external memory banks via intricately crafted controllers and hand-designed attentional mechanisms, our memory is internal, distributed, co-located alongside computation, and implicitly addressed, while being drastically simpler than prior efforts. Architecting networks with multigrid structure and connectivity, while distributing memory cells alongside computation throughout this topology, we observe the emergence of coherent memory subsystems. Our hierarchical spatial organization, parameterized convolutionally, permits efficient instantiation of large-capacity memories, while multigrid topology provides short internal routing pathways, allowing convolutional networks to efficiently approximate the behavior of fully connected networks. Such networks have an implicit capacity for internal attention; augmented with memory, they learn to read and write specific memory locations in a dynamic data-dependent manner. We demonstrate these capabilities on exploration and mapping tasks, where our network is able to self-organize and retain long-term memory for trajectories of thousands of time steps. On tasks decoupled from any notion of spatial geometry: sorting, associative recall, and question answering, our design functions as a truly generic memory and yields excellent results.

We propose a novel variant of SGD customized for training network architectures that support anytime behavior: such networks produce a series of increasingly accurate outputs over time. Efficient architectural designs for these networks focus on re-using internal state; subnetworks must produce representations relevant for both imme- diate prediction as well as refinement by subse- quent network stages. We consider traditional branched networks as well as a new class of re- cursively nested networks. Our new optimizer, Orthogonalized SGD, dynamically re-balances task-specific gradients when training a multitask network. In the context of anytime architectures, this optimizer projects gradients from later out- puts onto a parameter subspace that does not in- terfere with those from earlier outputs. Experi- ments demonstrate that training with Orthogonal- ized SGD significantly improves generalization accuracy of anytime networks.

The search for efficient, sparse deep neural network models is most prominently performed by pruning: training a dense, overparameterized network and removing parameters, usually via following a manually-crafted heuristic. Additionally, the recent Lottery Ticket Hypothesis conjectures that, for a typically-sized neural network, it is possible to find small sub-networks which, when trained from scratch on a comparable budget, match the performance of the original dense counterpart. We revisit fundamental aspects of pruning algorithms, pointing out missing ingredients in previous approaches, and develop a method, Continuous Sparsification, which searches for sparse networks based on a novel approximation of an intractable $\ell_0$ regularization. We compare against dominant heuristic-based methods on pruning as well as ticket search -- finding sparse subnetworks that can be successfully re-trained from an early iterate. Empirical results show that we surpass the state-of-the-art for both objectives, across models and datasets, including VGG trained on CIFAR-10 and ResNet-50 trained on ImageNet. In addition to setting a new standard for pruning, Continuous Sparsification also offers fast parallel ticket search, opening doors to new applications of the Lottery Ticket Hypothesis.

We create a framework for bootstrapping visual representation learning from a primitive visual grouping capability. We operationalize grouping via a contour detector that partitions an image into regions, followed by merging of those regions into a tree hierarchy. A small supervised dataset suffices for training this grouping primitive. Across a large unlabeled dataset, we apply this learned primitive to automatically predict hierarchical region structure. These predictions serve as guidance for self-supervised contrastive feature learning: we task a deep network with producing per-pixel embeddings whose pairwise distances respect the region hierarchy. Experiments demonstrate that our approach can serve as state-of-the-art generic pre-training, benefiting downstream tasks. We additionally explore applications to semantic region search and video-based object instance tracking.

We investigate and improve self-supervision as a drop-in replacement for ImageNet pretraining, focusing on automatic colorization as the proxy task. Self-supervised training has been shown to be more promising for utilizing unlabeled data than other, traditional unsupervised learning methods. We build on this success and evaluate the ability of our self-supervised network in several contexts. On VOC segmentation and classification tasks, we present results that are state-of-the-art among methods not using ImageNet labels for pretraining representations. Moreover, we present the first in-depth analysis of self-supervision via colorization, concluding that formulation of the loss, training details and network architecture play important roles in its effectiveness. This investigation is further expanded by revisiting the ImageNet pretraining paradigm, asking questions such as: How much training data is needed? How many labels are needed? How much do features change when fine-tuned? We relate these questions back to self-supervision by showing that colorization provides a similarly powerful supervisory signal as various flavors of ImageNet pretraining.

Spectral embedding provides a framework for solving perceptual organization problems, including image segmentation and figure/ground organization. From an affinity matrix describing pairwise relationships between pixels, it clusters pixels into regions, and, using a complex-valued extension, orders pixels according to layer. We train a convolutional neural network (CNN) to directly predict the pair-wise relationships that define this affinity matrix. Spectral embedding then resolves these predictions into a globally-consistent segmentation and figure/ground organization of the scene. Experiments demonstrate significant benefit to this direct coupling compared to prior works which use explicit intermediate stages, such as edge detection, on the pathway from image to affinities. Our results suggest spectral embedding as a powerful alternative to the conditional random field (CRF)-based globalization schemes typically coupled to deep neural networks.

We develop a new approach to inferring lightness, the perceived reflectance of surfaces, from a single image. Classic methods view this problem from the perspective of intrinsic image decomposition, where an image is separated into reflectance and shading components. Rather than reason about reflectance and shading together, we learn to directly predict lightness differences between pixels. Large-scale training from human judgement data on relative reflectance, and patch representations built using deep networks, provide the foundation for our model. Benchmarked on the Intrinsic Images in the Wild dataset [4], our local lightness model achieves on-par performance with the state-of-the-art global lightness model, which incorporates multiple shading/reflectance priors and simultaneous reasoning between pairs of pixels in a dense conditional random field formulation.

We introduce a new approach to intrinsic image decomposition, the task of decomposing a single image into albedo and shading components. Our strategy, which we term direct intrinsics, is to learn a convolutional neural network (CNN) that directly predicts output albedo and shading channels from an input RGB image patch. Direct intrinsics is a departure from classical techniques for intrinsic image decomposition, which typically rely on physically-motivated priors and graph-based inference algorithms. The large-scale synthetic ground-truth of the MPI Sintel dataset plays the key role in training direct intrinsics. We demonstrate results on both the synthetic images of Sintel and the real images of the classic MIT intrinsic image dataset. On Sintel, direct intrinsics, using only RGB input, outperforms all prior work, including methods that rely on RGB+Depth input. Direct intrinsics also generalizes across modalities, our Sintel-trained CNN produces quite reasonable decompositions on the real images of the MIT dataset. Our results indicate that the marriage of CNNs with synthetic training data may be a powerful new technique for tackling classic problems in computer vision.

We present a computer-assisted annotation system, together with a labeled dataset and benchmark suite, for evaluating an algorithm's ability to recover hierarchical scene structure.We evolve segmentation groundtruth from the two-dimensional image partition into a tree model that captures both occlusion and object-part relationships among possibly overlapping regions.Our tree model extends the segmentation problem to encompass object detection, object-part containment, and figure-ground ordering.We mitigate the cost of providing richer groundtruth labeling through a new webbased annotation tool with an intuitive graphical interface for rearranging the region hierarchy.Using precomputed superpixels, our tool also guides creation of user-specified regions with pixel-perfect boundaries.Widespread adoption of this human-machine combination should make the inaccuracies of bounding box labeling a relic of the past.Evaluating the state-of-the-art in fully automatic image segmentation reveals that it produces accurate two-dimension partitions, but does not respect groundtruth object-part structure.Our dataset and benchmark is the first to quantify these inadequacies.We illuminate recovery of rich scene structure as an important new goal for segmentation.

We reexamine the role of multiscale cues in image segmentation using an architecture that constructs a globally coherent scale-space output representation. This characteristic is in contrast to many existing works on bottom-up segmentation, which prematurely compress information into a single scale. The architecture is a standard extension of Normalized Cuts from an image plane to an image pyramid, with cross-scale constraints enforcing consistency in the solution while allowing emergence of coarse-to-fine detail. We observe that multiscale processing, in addition to improving segmentation quality, offers a route by which to speed computation. We make a significant algorithmic advance in the form of a custom multigrid eigensolver for constrained Angular Embedding problems possessing coarse-to-fine structure. Multiscale Normalized Cuts is a special case. Our solver builds atop recent results on randomized matrix approximation, using a novel interpolation operation to mold its computational strategy according to cross-scale constraints in the problem definition. Applying our solver to multiscale segmentation problems demonstrates speedup by more than an order of magnitude. This speedup is at the algorithmic level and carries over to any implementation target.

In this work, we propose a contour and region detector for video data that exploits motion cues and distinguishes occlusion boundaries from internal boundaries based on optical flow. This detector outperforms the state-of-the-art on the benchmark of Stein and Hebert, improving average precision from .58 to .72. Moreover, the optical flow on and near occlusion boundaries allows us to assign a depth ordering to the adjacent regions. To evaluate performance on this edge-based figure/ground labeling task, we introduce a new video dataset that we believe will support further research in the field by allowing quantitative comparison of computational models for occlusion boundary detection, depth ordering and segmentation in video sequences.

We present a new framework in which image segmentation, figure/ground organization, and object detection all appear as the result of solving a single grouping problem. This framework serves as a perceptual organization stage that integrates information from low-level image cues with that of high-level part detectors. Pixels and parts each appear as nodes in a graph whose edges encode both affinity and ordering relationships. We derive a generalized eigen-problem from this graph and read off an interpretation of the image from the solution eigenvectors. Combining an off-the-shelf top-down part-based person detector with our low-level cues and grouping formulation, we demonstrate improvements to object detection and segmentation.

This paper investigates two fundamental problems in computer vision: contour detection and image segmentation. We present state-of-the-art algorithms for both of these tasks. Our contour detector combines multiple local cues into a globalization framework based on spectral clustering. Our segmentation algorithm consists of generic machinery for transforming the output of any contour detector into a hierarchical region tree. In this manner, we reduce the problem of image segmentation to that of contour detection. Extensive experimental evaluation demonstrates that both our contour detection and segmentation methods significantly outperform competing algorithms. The automatically generated hierarchical segmentations can be interactively refined by user-specified annotations. Computation at multiple image resolutions provides a means of coupling our system to recognition applications.

We propose a generic grouping algorithm that constructs a hierarchy of regions from the output of any contour detector. Our method consists of two steps, an oriented watershed transform (OWT) to form initial regions from contours, followed by construction of an ultra-metric contour map (UCM) defining a hierarchical segmentation. We provide extensive experimental evaluation to demonstrate that, when coupled to a high-performance contour detector, the OWT-UCM algorithm produces state-of-the-art image segmentations. These hierarchical segmentations can optionally be further refined by user-specified annotations.

Contours and junctions are important cues for perceptual organization and shape recognition. Detecting junctions locally has proved problematic because the image intensity surface is confusing in the neighborhood of a junction. Edge detectors also do not perform well near junctions. Current leading approaches to junction detection, such as the Harris operator, are based on 2D variation in the intensity signal. However, a drawback of this strategy is that it confuses textured regions with junctions. We believe that the right approach to junction detection should take advantage of the contours that are incident at a junction; contours themselves can be detected by processes that use more global approaches. In this paper, we develop a new high-performance contour detector using a combination of local and global cues. This contour detector provides the best performance to date (F=0.70) on the Berkeley Segmentation Dataset (BSDS) benchmark. From the resulting contours, we detect and localize candidate junctions, taking into account both contour salience and geometric configuration. We show that improvements in our contour model lead to better junctions. Our contour and junction detectors both provide state of the art performance.

We consider visual category recognition in the framework of measuring similarities, or equivalently perceptual distances, to prototype examples of categories. This approach is quite flexible, and permits recognition based on color, texture, and particularly shape, in a homogeneous framework. While nearest neighbor classifiers are natural in this setting, they suffer from the problem of high variance (in bias-variance decomposition) in the case of limited sampling. Alternatively, one could use support vector machines but they involve time-consuming optimization and computation of pairwise distances. We propose a hybrid of these two methods which deals naturally with the multiclass setting, has reasonable computational complexity both in training and at run time, and yields excellent results in practice. The basic idea is to find close neighbors to a query sample and train a local support vector machine that preserves the distance function on the collection of neighbors. Our method can be applied to large, multiclass data sets for which it outperforms nearest neighbor and support vector machines, and remains efficient when the problem becomes intractable for support vector machines. A wide variety of distance functions can be used and our experiments show state-of-the-art performance on a number of benchmark data sets for shape and texture classification (MNIST, USPS, CUReT) and object recognition (Caltech- 101). On Caltech-101 we achieved a correct classification rate of 59.05%(±0.56%) at 15 training images per class, and 66.23%(±0.48%) at 30 training images.

We describe a method that can make a scanned, handwritten mediaeval latin manuscript accessible to full text search. A generalized HMM is fitted, using transcribed latin to obtain a transition model and one exam- ple each of 22 letters to obtain an emission model. We show results for unigram, bigram and trigram models. Our method transcribes 25 pages of a manuscript of Terence with fair accuracy (75% of letters correctly transcribed). Search results are very strong; we use examples of vari- ant spellings to demonstrate that the search respects the ink of the doc- ument. Furthermore, our model produces fair searches on a document from which we obtained no training data.