While image-text representation learning has become very popular in recent years, existing models tend to lack spatial awareness and have limited direct applicability for dense understanding tasks. For this reason, self-supervised image-only pretraining is still the go-to method for many dense vision applications (e.g. depth estimation, semantic segmentation), despite the lack of explicit supervisory signals. In this paper, we close this gap between image-text and self-supervised learning, by proposing a novel general-purpose image-text model, which can be effectively used off the shelf for dense and global vision tasks. Our method, which we refer to as Text-Image Pretraining with Spatial awareness (TIPS), leverages two simple and effective insights. First, on textual supervision: we reveal that replacing noisy web image captions by synthetically generated textual descriptions boosts dense understanding performance significantly, due to a much richer signal for learning spatially aware representations. We propose an adapted training method that combines noisy and synthetic captions, resulting in improvements across both dense and global understanding tasks. Second, on the learning technique: we propose to combine contrastive image-text learning with self-supervised masked image modeling, to encourage spatial coherence, unlocking substantial enhancements for downstream applications. Building on these two ideas, we scale our model using the transformer architecture, trained on a curated set of public images. Our experiments are conducted on $8$ tasks involving $16$ datasets in total, demonstrating strong off-the-shelf performance on both dense and global understanding, for several image-only and image-text tasks. Code and models are released at https://github.com/google-deepmind/tips .

We explore the boundaries of scaling up a multilingual vision and language model, both in terms of size of the components and the breadth of its training task mixture. Our model achieves new levels of performance on a wide-range of varied and complex tasks, including multiple image-based captioning and question-answering tasks, image-based document understanding and few-shot (in-context) learning, as well as object detection, video question answering, and video captioning. Our model advances the state-of-the-art on most vision-and-language benchmarks considered (20+ of them). Finally, we observe emerging capabilities, such as complex counting and multilingual object detection, tasks that are not explicitly in the training mix.

Fine-grained and instance-level recognition methods are commonly trained and evaluated on specific domains, in a model per domain scenario. Such an approach, however, is impractical in real large-scale applications. In this work, we address the problem of universal image embedding, where a single universal model is trained and used in multiple domains. First, we leverage existing domain-specific datasets to carefully construct a new large-scale public benchmark for the evaluation of universal image embeddings, with 241k query images, 1.4M index images and 2.8M training images across 8 different domains and 349k classes. We define suitable metrics, training and evaluation protocols to foster future research in this area. Second, we provide a comprehensive experimental evaluation on the new dataset, demonstrating that existing approaches and simplistic extensions lead to worse performance than an assembly of models trained for each domain separately. Finally, we conducted a public research competition on this topic, leveraging industrial datasets, which attracted the participation of more than 1k teams world-wide. This exercise generated many interesting research ideas and findings which we present in detail. Project webpage: https://cmp.felk.cvut.cz/univ_emb/

Semantic segmentation is the problem of assigning an object label to each pixel. It unifies the image segmentation and object recognition problems. The importance of using contextual information in semantic segmentation frameworks has been widely realized in the field. We propose a contextual framework, called contextual hierarchical model (CHM), which learns contextual information in a hierarchical framework for semantic segmentation. At each level of the hierarchy, a classifier is trained based on downsampled input images and outputs of previous levels. Our model then incorporates the resulting multi-resolution contextual information into a classifier to segment the input image at original resolution. This training strategy allows for optimization of a joint posterior probability at multiple resolutions through the hierarchy. Contextual hierarchical model is purely based on the input image patches and does not make use of any fragments or shape examples. Hence, it is applicable to a variety of problems such as object segmentation and edge detection. We demonstrate that CHM performs at par with state-of-the-art on Stanford background and Weizmann horse datasets. It also outperforms state-of-the-art edge detection methods on NYU depth dataset and achieves state-of-the-art on Berkeley segmentation dataset (BSDS 500).

This paper investigates one of the most fundamental computer vision problems: image segmentation. We propose a supervised hierarchical approach to object-independent image segmentation. Starting with oversegmenting superpixels, we use a tree structure to represent the hierarchy of region merging, by which we reduce the problem of segmenting image regions to finding a set of label assignment to tree nodes. We formulate the tree structure as a constrained conditional model to associate region merging with likelihoods predicted using an ensemble boundary classifier. Final segmentations can then be inferred by finding globally optimal solutions to the model efficiently. We also present an iterative training and testing algorithm that generates various tree structures and combines them to emphasize accurate boundaries by segmentation accumulation. Experiment results and comparisons with other recent methods on six public data sets demonstrate that our approach achieves the state-of-the-art region accuracy and is competitive in image segmentation without semantic priors.

We introduce a novel general regression model that is based on a linear combination of a new set of non-local basis functions that forms an effective feature space. We propose a training algorithm that learns all the model parameters simultaneously and offer an initialization scheme for parameters of the basis functions. We show through several experiments that the proposed method offers better coverage for high-dimensional space compared to local Gaussian basis functions and provides competitive performance in comparison to other state-of-the-art regression methods.

Contextual information plays an important role in solving vision problems such as image segmentation. However, extracting contextual information and using it in an effective way remains a difficult problem. To address this challenge, we propose a multi-resolution contextual framework, called cascaded hierarchical model (CHM), which learns contextual information in a hierarchical framework for image segmentation. At each level of the hierarchy, a classifier is trained based on downsampled input images and outputs of previous levels. Our model then incorporates the resulting multi-resolution contextual information into a classifier to segment the input image at original resolution. We repeat this procedure by cascading the hierarchical framework to improve the segmentation accuracy. Multiple classifiers are learned in the CHM; therefore, a fast and accurate classifier is required to make the training tractable. The classifier also needs to be robust against overfitting due to the large number of parameters learned during training. We introduce a novel classification scheme, called logistic disjunctive normal networks (LDNN), which consists of one adaptive layer of feature detectors implemented by logistic sigmoid functions followed by two fixed layers of logical units that compute conjunctions and disjunctions, respectively. We demonstrate that LDNN outperforms state-of-theart classifiers and can be used in the CHM to improve object segmentation performance.

Automated electron microscopy (EM) image analysis techniques can be tremendously helpful for connectomics research. In this paper, we extend our previous work [1] and propose a fully automatic method to utilize inter-section information for intra-section neuron segmentation of EM image stacks. A watershed merge forest is built via the watershed transform with each tree representing the region merging hierarchy of one 2D section in the stack. A section classifier is learned to identify the most likely region correspondence between adjacent sections. The inter-section information from such correspondence is incorporated to update the potentials of tree nodes. We resolve the merge forest using these potentials together with consistency constraints to acquire the final segmentation of the whole stack. We demonstrate that our method leads to notable segmentation accuracy improvement by experimenting with two types of EM image data sets.

Contextual information has been widely used as a rich source of information to segment multiple objects in an image. A contextual model uses the relationships between the objects in a scene to facilitate object detection and segmentation. Using contextual information from different objects in an effective way for object segmentation, however, remains a difficult problem. In this paper, we introduce a novel framework, called multiclass multiscale (MCMS) series contextual model, which uses contextual information from multiple objects and at different scales for learning discriminative models in a supervised setting. The MCMS model incorporates cross-object and inter-object information into one probabilistic framework and thus is able to capture geometrical relationships and dependencies among multiple objects in addition to local information from each single object present in an image. We demonstrate that our MCMS model improves object segmentation performance in electron microscopy images and provides a coherent segmentation of multiple objects. Through speeding up the segmentation process, the proposed method will allow neurobiologists to move beyond individual specimens and analyze populations paving the way for understanding neurodegenerative diseases at the microscopic level.

In this paper, a new AdaBoost learning framework, called WNS-AdaBoost, is proposed for training discriminative models. The proposed approach significantly speeds up the learning process of adaptive boosting (AdaBoost) by reducing the number of data points. For this purpose, we introduce the weighted novelty selection (WNS) sampling strategy and combine it with AdaBoost to obtain an efficient and fast learning algorithm. WNS selects a representative subset of data thereby reducing the number of data points onto which AdaBoost is applied. In addition, WNS associates a weight with each selected data point such that the weighted subset approximates the distribution of all the training data. This ensures that AdaBoost can trained efficiently and with minimal loss of accuracy. The performance of WNS-AdaBoost is first demonstrated in a classification task. Then, WNS is employed in a probabilistic boosting-tree (PBT) structure for image segmentation. Results in these two applications show that the training time using WNS-AdaBoost is greatly reduced at the cost of only a few percent in accuracy.

We propose a three-state series neural network for effective propagation of context and uncertainty information for image parsing. The activation functions used in the proposed model have three states instead of the normal two states. This makes the neural network more flexible than the two-state neural network, and allows for uncertainty to be propagated through the stages. In other words, decisions about difficult pixels can be left for later stages which have access to more contextual information than earlier stages. We applied the proposed method to three different datasets and experimental results demonstrate higher performance of the three-state series neural network.