We introduce $\textbf{STRING}$: 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 $\textbf{exact}$ 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. Videos of STRING-based robotics controllers can be found here: https://sites.google.com/view/string-robotics.

The most performant spatio-temporal action localisation models use external person proposals and complex external memory banks. We propose a fully end-to-end, purely-transformer based model that directly ingests an input video, and outputs tubelets - a sequence of bounding boxes and the action classes at each frame. Our flexible model can be trained with either sparse bounding-box supervision on individual frames, or full tubelet annotations. And in both cases, it predicts coherent tubelets as the output. Moreover, our end-to-end model requires no additional pre-processing in the form of proposals, or post-processing in terms of non-maximal suppression. We perform extensive ablation experiments, and significantly advance the state-of-the-art on five different spatio-temporal action localisation benchmarks with both sparse keyframes and full tubelet annotations.

As foundation models become more popular, there is a growing need to efficiently finetune them for downstream tasks. Although numerous adaptation methods have been proposed, they are designed to be efficient only in terms of how many parameters are trained. They, however, typi-cally still require backpropagating gradients throughout the model, meaning that their training-time and -memory cost does not reduce as significantly. We propose an adaptation method which does not back-propagate gradients through the backbone. We achieve this by designing a lightweight network in parallel that oper-ates on features from the frozen, pretrained backbone. As a result, our method is efficient not only in terms of parame-ters, but also in training-time and memory usage. Our approach achieves state-of-the-art accuracy-parameter trade-offs on the popular VTAB benchmark, and we further show how we outperform prior works with respect to training-time and -memory usage too. We further demonstrate the training efficiency and scalability of our method by adapting a vision transformer backbone of 4 billion parameters for the computationally demanding task of video classifi-cation, without any intricate model parallelism. Here, we outperform a prior adaptor-based method which could only scale to a 1 billion parameter backbone, or fully-finetuning a smaller backbone, with the same GPU and less training time. To facilitate further research, we release code at https://github.com/google-researchlscenic.

We introduce SPARse fine-grained Contrastive alignment (SPARC), a simple method for pretraining more fine-grained multimodal representations from image-text pairs. Given that multiple image patches often correspond to single words, we propose to learn a grouping of image patches for every token in the caption. To achieve this, we use a sparse similarity metric between image patches and language tokens and compute for each token a language-grouped vision embedding as the weighted average of patches. The token and language-grouped vision embeddings are then contrasted through a fine-grained sequence-wise loss that only depends on individual samples and does not require other batch samples as negatives, i.e., more detailed information is encoded in a computationally inexpensive way. SPARC combines this fine-grained loss with a contrastive loss between global image and text embeddings to learn representations that simultaneously encode global and local information. We thoroughly evaluate SPARC and show improved performance over competing approaches both on image-level tasks relying on coarse-grained information, e.g. classification, as well as region-level tasks relying on fine-grained information, e.g., retrieval, object detection, segmentation while also improving model faithfulness and captioning in foundational vision-language models.

We present an architecture and a training recipe that adapts pretrained 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 pretraining, can be transferred successfully to open-world localization across diverse videos.

The scaling of Transformers has driven breakthrough capabilities for language models. At present, the largest large language models (LLMs) contain upwards of 100B parameters. Vision Transformers (ViT) have introduced the same architecture to image and video modelling, but these have not yet been successfully scaled to nearly the same degree; the largest dense ViT contains 4B parameters (Chen et al., 2022). We present a recipe for highly efficient and stable training of a 22B-parameter ViT (ViT-22B) and perform a wide variety of experiments on the resulting model. When evaluated on downstream tasks (often with a lightweight linear model on frozen features), ViT-22B demonstrates increasing performance with scale. We further observe other interesting benefits of scale, including an improved tradeoff between fairness and performance, state-of-the-art alignment to human visual perception in terms of shape/texture bias, and improved robustness. ViT-22B demonstrates the potential for "LLM-like" scaling in vision, and provides key steps towards getting there.

The ubiquitous and demonstrably suboptimal choice of resizing images to a fixed resolution before processing them with computer vision models has not yet been successfully challenged. However, models such as the Vision Transformer (ViT) offer flexible sequence-based modeling, and hence varying input sequence lengths. We take advantage of this with NaViT (Native Resolution ViT) which uses sequence packing during training to process inputs of arbitrary resolutions and aspect ratios. Alongside flexible model usage, we demonstrate improved training efficiency for large-scale supervised and contrastive image-text pretraining. NaViT can be efficiently transferred to standard tasks such as image and video classification, object detection, and semantic segmentation and leads to improved results on robustness and fairness benchmarks. At inference time, the input resolution flexibility can be used to smoothly navigate the test-time cost-performance trade-off. We believe that NaViTmarks a departure from the standard, CNN-designed, input and modelling pipeline used by most computer vision models, and represents a promising direction for ViTs.

Open-vocabulary object detection has benefited greatly from pretrained vision-language models, but is still limited by the amount of available detection training data. While detection training data can be expanded by using Web image-text pairs as weak supervision, this has not been done at scales comparable to image-level pretraining. Here, we scale up detection data with self-training, which uses an existing detector to generate pseudo-box annotations on image-text pairs. Major challenges in scaling self-training are the choice of label space, pseudo-annotation filtering, and training efficiency. We present the OWLv2 model and OWL-ST self-training recipe, which address these challenges. OWLv2 surpasses the performance of previous state-of-the-art open-vocabulary detectors already at comparable training scales (~10M examples). However, with OWL-ST, we can scale to over 1B examples, yielding further large improvement: With an L/14 architecture, OWL-ST improves AP on LVIS rare classes, for which the model has seen no human box annotations, from 31.2% to 44.6% (43% relative improvement). OWL-ST unlocks Web-scale training for open-world localization, similar to what has been seen for image classification and language modelling. Code and checkpoints are available on GitHub.

Scenic is an open-source11https://github.com/google-research/scenic JAX library with a focus on transformer-based models for computer vision research and beyond. The goal of this toolkit is to facilitate rapid experimentation, prototyping, and research of new architectures and models. Scenic supports a diverse range of tasks (e.g., classification, segmentation, detection) and facilitates working on multi-modal problems, along with GPU/TPU support for large-scale, multi-host and multi-device training. Scenic also offers optimized implementations of state-of-the-art research models spanning a wide range of modalities. Scenic has been successfully used for numerous projects and published papers and continues serving as the library of choice for rapid prototyping and publication of new research ideas.

Generating temporally coherent high fidelity video is an important milestone in generative modeling research. We make progress towards this milestone by proposing a diffusion model for video generation that shows very promising initial results. Our model is a natural extension of the standard image diffusion architecture, and it enables jointly training from image and video data, which we find to reduce the variance of minibatch gradients and speed up optimization. To generate long and higher resolution videos we introduce a new conditional sampling technique for spatial and temporal video extension that performs better than previously proposed methods. We present the first results on a large text-conditioned video generation task, as well as state-of-the-art results on established benchmarks for video prediction and unconditional video generation. Supplementary material is available at https://video-diffusion.github.io/.

Speech synthesis is an important practical generative modeling problem that has seen great progress over the last few years, with likelihood-based autoregressive neural models now outperforming traditional concatenative systems. A downside of such autoregressive models is that they require executing tens of thousands of sequential operations per second of generated audio, making them ill-suited for deployment on specialized deep learning hardware. Here, we propose a new learning method that allows us to train highly parallel models of speech, without requiring access to an analytical likelihood function. Our approach is based on a generalized energy distance between the distributions of the generated and real audio. This spectral energy distance is a proper scoring rule with respect to the distribution over magnitude-spectrograms of the generated waveform audio and offers statistical consistency guarantees. The distance can be calculated from minibatches without bias, and does not involve adversarial learning, yielding a stable and consistent method for training implicit generative models. Empirically, we achieve state-of-the-art generation quality among implicit generative models, as judged by the recently-proposed cFDSD metric. When combining our method with adversarial techniques, we also improve upon the recently-proposed GAN-TTS model in terms of Mean Opinion Score as judged by trained human evaluators.

Translation of mRNAs through Internal Ribosome Entry Sites (IRESs) has emerged as a prominent mechanism of cellular and viral initiation. It supports cap-independent translation of select cellular genes under normal conditions, and in conditions when cap-dependent translation is inhibited. IRES structure and sequence are believed to be involved in this process. However due to the small number of IRESs known, there have been no systematic investigations of the determinants of IRES activity. With the recent discovery of thousands of novel IRESs in human and viruses, the next challenge is to decipher the sequence determinants of IRES activity. We present the first in-depth computational analysis of a large body of IRESs, exploring RNA sequence features predictive of IRES activity. We identified predictive k-mer features resembling IRES trans-acting factor (ITAF) binding motifs across human and viral IRESs, and found that their effect on expression depends on their sequence, number and position. Our results also suggest that the architecture of retroviral IRESs differs from that of other viruses, presumably due to their exposure to the nuclear environment. Finally, we measured IRES activity of synthetically designed sequences to confirm our prediction of increasing activity as a function of the number of short IRES elements.

Translation of RNA to protein is a core process for any living organism. While for some steps of this process the effect on protein production is understood, a holistic understanding of translation still remains elusive. In silico modelling is a promising approach for elucidating the process of protein synthesis. Although a number of computational models of the process have been proposed, their application is limited by the assumptions they make. Ribosome profiling (RP), a relatively new sequencing-based technique capable of recording snapshots of the locations of actively translating ribosomes, is a promising source of information for deriving unbiased data-driven translation models. However, quantitative analysis of RP data is challenging due to high measurement variance and the inability to discriminate between the number of ribosomes measured on a gene and their speed of translation. We propose a solution in the form of a novel multi-scale interpretation of RP data that allows for deriving models with translation dynamics extracted from the snapshots. We demonstrate the usefulness of this approach by simultaneously determining for the first time per-codon translation elongation and per-gene translation initiation rates of Saccharomyces cerevisiae from RP data for two versions of the Totally Asymmetric Exclusion Process (TASEP) model of translation. We do this in an unbiased fashion, by fitting the models using only RP data with a novel optimization scheme based on Monte Carlo simulation to keep the problem tractable. The fitted models match the data significantly better than existing models and their predictions show better agreement with several independent protein abundance datasets than existing models. Results additionally indicate that the tRNA pool adaptation hypothesis is incomplete, with evidence suggesting that tRNA post-transcriptional modifications and codon context may play a role in determining codon elongation rates.

MOTIVATION: The increasing availability of second-generation high-throughput sequencing (HTS) technologies has sparked a growing interest in de novo genome sequencing. This in turn has fueled the need for reliable means of obtaining high-quality draft genomes from short-read sequencing data. The millions of reads usually involved in HTS experiments are first assembled into longer fragments called contigs, which are then scaffolded, i.e. ordered and oriented using additional information, to produce even longer sequences called scaffolds. Most existing scaffolders of HTS genome assemblies are not suited for using information other than paired reads to perform scaffolding. They use this limited information to construct scaffolds, often preferring scaffold length over accuracy, when faced with the tradeoff. RESULTS: We present GRASS (GeneRic ASsembly Scaffolder)-a novel algorithm for scaffolding second-generation sequencing assemblies capable of using diverse information sources. GRASS offers a mixed-integer programming formulation of the contig scaffolding problem, which combines contig order, distance and orientation in a single optimization objective. The resulting optimization problem is solved using an expectation-maximization procedure and an unconstrained binary quadratic programming approximation of the original problem. We compared GRASS with existing HTS scaffolders using Illumina paired reads of three bacterial genomes. Our algorithm constructs a comparable number of scaffolds, but makes fewer errors. This result is further improved when additional data, in the form of related genome sequences, are used. AVAILABILITY: GRASS source code is freely available from http://code.google.com/p/tud-scaffolding/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.