In this work, we demonstrate that due to the inadequacies in the existing evaluation protocols and datasets, there is a need to revisit and comprehensively examine the multimodal Zero-Shot Learning (MZSL) problem formulation. Specifically, we address two major challenges faced by current MZSL approaches; (1) Established baselines are frequently incomparable and occasionally even flawed since existing evaluation datasets often have some overlap with the training dataset, thus violating the zero-shot paradigm; (2) Most existing methods are biased towards seen classes, which significantly reduces the performance when evaluated on both seen and unseen classes. To address these challenges, we first introduce a new multimodal dataset for zero-shot evaluation called MZSL-50 with 4462 videos from 50 widely diversified classes and no overlap with the training data. Further, we propose a novel multimodal zero-shot transformer (MZST) architecture that leverages attention bottlenecks for multimodal fusion. Our model directly predicts the semantic representation and is superior at reducing the bias towards seen classes. We conduct extensive ablation studies, and achieve state-of-the-art results on three benchmark datasets and our novel MZSL-50 dataset. Specifically, we improve the conventional MZSL performance by a margin of 2.1%, 9.81% and 8.68% on VGG-Sound, UCF-101 and ActivityNet, respectively. Finally, we expect the introduction of the MZSL-50 dataset will promote the future in-depth research on multimodal zero-shot learning in the community.1

Grounding-based vision and language models have been successfully applied to low-level vision tasks, aiming to precisely locate objects referred in captions. The effectiveness of grounding representation learning heavily relies on the scale of the training dataset. Despite being a useful data enrichment strategy, data augmentation has received minimal attention in existing vision and language tasks as augmentation for image-caption pairs is non-trivial. In this study, we propose a robust phrase grounding model trained with text-conditioned and text-unconditioned data augmentations. Specifically, we apply text-conditioned color jittering and horizontal flipping to ensure semantic consistency between images and captions. To guarantee image-caption correspondence in the training samples, we modify the captions according to pre-defined keywords when applying horizontal flipping. Additionally, inspired by recent masked signal reconstruction, we propose to use pixel-level masking as a novel form of data augmentation. While we demonstrate our data augmentation method with MDETR framework, the proposed approach is applicable to common grounding-based vision and language tasks with other frameworks. Finally, we show that image encoder pretrained on large-scale image and language datasets (such as CLIP) can further improve the results. Through extensive experiments on three commonly applied datasets: Flickr30k, referring expressions and GQA, our method demonstrates advanced performance over the state-of-the-arts with various metrics. Code can be found in https://github.com/amzn/augment-the-pairs-wacv2024.

The recently proposed Vision transformers (ViTs) have shown very impressive empirical performance in various computer vision tasks, and they are viewed as an important type of foundation model. However, ViTs are typically constructed with large-scale sizes, which then severely hinder their potential deployment in many practical resources constrained applications. To mitigate this challenging problem, structured pruning is a promising solution to compress model size and enable practical efficiency. However, unlike its current popularity for CNNs and RNNs, structured pruning for ViT models is little explored. In this paper, we propose GOHSP, a unified framework of Graph and Optimization-based Structured Pruning for ViT models. We first develop a graph-based ranking for measuring the importance of attention heads, and the extracted importance information is further integrated to an optimization-based procedure to impose the heterogeneous structured sparsity patterns on the ViT models. Experimental results show that our proposed GOHSP demonstrates excellent compression performance. On CIFAR-10 dataset, our approach can bring 40% parameters reduction with no accuracy loss for ViT-Small model. On ImageNet dataset, with 30% and 35% sparsity ratio for DeiT-Tiny and DeiT-Small models, our approach achieves 1.65% and 0.76% accuracy increase over the existing structured pruning methods, respectively.

Transformers have been recently utilized for vision and language tasks successfully. For example, recent image and language models with more than 200M parameters have been proposed to learn visual grounding in the pre-training step and show impressive results on downstream vision and language tasks. On the other hand, there exists a large amount of computational redundancy in these large models which skips their run-time efficiency. To address this problem, we propose dynamic inference for grounding based vision and language models conditioned on the input image-text pair. We first design an approach to dynamically skip multihead self-attention and feed forward network layers across two backbones and multimodal network. Additionally, we propose dynamic token pruning and fusion for two backbones. In particular, we remove redundant tokens at different levels of the backbones and fuse the image tokens with the language tokens in an adaptive manner. To learn policies for dynamic inference, we train agents using reinforcement learning. In this direction, we replace the CNN backbone in a recent grounding-based vision and language model, MDETR, with a vision transformer and call it ViTMDETR. Then, we apply our dynamic inference method to ViTMDETR, called D-ViTDMETR, and perform experiments on image-language tasks. Our results show that we can improve the run-time efficiency of the state-of-the-art models MDETR and GLIP by up to ~ 50% on Referring Expression Comprehension and Segmentation, and VQA with only maximum ~ 0.3% accuracy drop.

Recent multi-modal detectors based on transformers and modality encoders have successfully achieved impressive results on end-to-end visual object detection conditioned on a raw text query. However, they require a large model size and an enormous amount of computations to achieve high performance, which makes it difficult to deploy mobile applications that are limited by tight hardware resources. In this paper, we present a Lightweight modulated detector, Lite-MDETR, to facilitate efficient end-to-end multi-modal understanding on mobile devices. The key primitive is that Dictionary-Lookup-Transformormations (DLT) is proposed to replace Linear Transformation (LT) in multi-modal detectors where each weight in Linear Transformation (LT) is approximately factorized into a smaller dictionary, index, and coefficient. This way, the enormous linear projection with weights is converted into efficient linear projection with dictionaries, a few lookups and scalings with indices and coefficients. DLT can be applied to any pretrained multi-modal detectors, removing the need to perform expensive training from scratch. To tackle the challenging training of DLT due to non-differentiable index, we convert the index and coefficient into a sparse matrix, train this sparse matrix during the fine-tuning phase, and recover it back to index and coefficient during the inference phase. Our experiments on phrase grounding, referring expression comprehension and segmentation, and VQA show that our Lite-MDETR achieves similar accuracy as the prior multi-modal detectors with up to ~ 4.1 × model size reduction.

The combination of high-resolution satellite imagery and machine learning have proven useful in many sustainability-related tasks, including poverty prediction, infrastructure measurement, and forest monitoring. However, the accuracy afforded by high-resolution imagery comes at a cost, as such imagery is extremely expensive to purchase at scale. This creates a substantial hurdle to the efficient scaling and widespread adoption of high-resolution-based approaches. To reduce acquisition costs while maintaining accuracy, we propose a reinforcement learning approach in which free low-resolution imagery is used to dynamically identify where to acquire costly high-resolution images, prior to performing a deep learning task on the high-resolution images. We apply this approach to the task of poverty prediction in Uganda, building on an earlier approach that used object detection to count objects and use these counts to predict poverty. Our approach exceeds previous performance benchmarks on this task while using 80% fewer high-resolution images, and could be useful in many domains that require high-resolution imagery.

Major decisions from governments and other large organizations rely on measurements of the populace's well-being, but making such measurements at a broad scale is expensive and thus infrequent in much of the developing world. We propose an inexpensive, scalable, and interpretable approach to predict key livelihood indicators from public crowd-sourced street-level imagery. Such imagery can be cheaply collected and more frequently updated compared to traditional surveying methods, while containing plausibly relevant information for a range of livelihood indicators. We propose two approaches to learn from the street-level imagery: (1) a method that creates multi-household cluster representations by detecting informative objects and (2) a graph-based approach that captures the relationships between images. By visualizing what features are important to a model and how they are used, we can help end-user organizations understand the models and offer an alternate approach for index estimation that uses cheaply obtained roadway features. By comparing our results against ground data collected in nationally-representative household surveys, we demonstrate the performance of our approach in accurately predicting indicators of poverty, population, and health and its scalability by testing in two different countries, India and Kenya. Our code is available at https://github.com/sustainlab-group/mapillarygcn.

Contrastive learning methods have significantly narrowed the gap between supervised and unsupervised learning on computer vision tasks. In this paper, we explore their application to geo-located datasets, e.g. remote sensing, where unlabeled data is often abundant but labeled data is scarce. We first show that due to their different characteristics, a non-trivial gap persists between contrastive and supervised learning on standard benchmarks. To close the gap, we propose novel training methods that exploit the spatio-temporal structure of remote sensing data. We leverage spatially aligned images over time to construct temporal positive pairs in contrastive learning and geo-location to design pre-text tasks. Our experiments show that our proposed method closes the gap between contrastive and supervised learning on image classification, object detection and semantic segmentation for remote sensing. Moreover, we demonstrate that the proposed method can also be applied to geo-tagged ImageNet images, improving downstream performance on various tasks.

While high resolution images contain semantically more useful information than their lower resolution counterparts, processing them is computationally more expensive, and in some applications, e.g. remote sensing, they can be much more expensive to acquire. For these reasons, it is desirable to develop an automatic method to selectively use high resolution data when necessary while maintaining accuracy and reducing acquisition/run-time cost. In this direction, we propose PatchDrop a reinforcement learning approach to dynamically identify when and where to use/acquire high resolution data conditioned on the paired, cheap, low resolution images. We conduct experiments on CIFAR10, CIFAR100, ImageNet and fMoW datasets where we use significantly less high resolution data while maintaining similar accuracy to models which use full high resolution images.

Accurate local-level poverty measurement is an essential task for governments and humanitarian organizations to track the progress towards improving livelihoods and distribute scarce resources. Recent computer vision advances in using satellite imagery to predict poverty have shown increasing accuracy, but they do not generate features that are interpretable to policymakers, inhibiting adoption by practitioners. Here we demonstrate an interpretable computational framework to accurately predict poverty at a local level by applying object detectors to high resolution (30cm) satellite images. Using the weighted counts of objects as features, we achieve 0.539 Pearson's r^2 in predicting village-level poverty in Uganda, a 31% improvement over existing (and less interpretable) benchmarks. Feature importance and ablation analysis reveal intuitive relationships between object counts and poverty predictions. Our results suggest that interpretability does not have to come at the cost of performance, at least in this important domain.

Satellite images hold great promise for continuous environmental monitoring and earth observation. Occlusions cast by clouds, however, can severely limit coverage, making ground information extraction more difficult. Existing pipelines typically perform cloud removal with simple temporal composites and hand-crafted filters. In contrast, we cast the problem of cloud removal as a conditional image synthesis challenge, and we propose a trainable spatiotemporal generator network (STGAN) to remove clouds. We train our model on a new large-scale spatiotemporal dataset that we construct, containing 97640 image pairs covering all continents. We demonstrate experimentally that the proposed STGAN model outperforms standard models and can generate realistic cloud-free images with high PSNR and SSIM values across a variety of atmospheric conditions, leading to improved performance in downstream tasks such as land cover classification.

Traditionally, an object detector is applied to every part of the scene of interest, and its accuracy and computational cost increases with higher resolution images. However, in some application domains such as remote sensing, purchasing high spatial resolution images is expensive. To reduce the large computational and monetary cost associated with using high spatial resolution images, we propose a reinforcement learning agent that adaptively selects the spatial resolution of each image that is provided to the detector. In particular, we train the agent in a dual reward setting to choose low spatial resolution images to be run through a coarse level detector when the image is dominated by large objects, and high spatial resolution images to be run through a fine level detector when it is dominated by small objects. This reduces the dependency on high spatial resolution images for building a robust detector and increases run-time efficiency. We perform experiments on the xView dataset, consisting of large images, where we increase runtime efficiency by 50% and use high resolution images only 30% of the time while maintaining similar accuracy as a detector that uses only high resolution images.

Despite recent progress in computer vision, fine-grained interpretation of satellite images remains challenging because of a lack of labeled training data. To overcome this limitation, we construct a novel dataset called WikiSatNet by pairing geo-referenced Wikipedia articles with satellite imagery of their corresponding locations. We then propose two strategies to learn representations of satellite images by predicting properties of the corresponding articles from the images. Leveraging this new multi-modal dataset, we can drastically reduce the quantity of human-annotated labels and time required for downstream tasks. On the recently released fMoW dataset, our pre-training strategies can boost the performance of a model pre-trained on ImageNet by up to 4.5% in F1 score.

Progress on the UN Sustainable Development Goals (SDGs) is hampered by a persistent lack of data regarding key social, environmental, and economic indicators, particularly in developing countries. For example, data on poverty - the first of seventeen SDGs - is both spatially sparse and infrequently collected in Sub-Saharan Africa due to the high cost of surveys. Here we propose a novel method for estimating socioeconomic indicators using open-source, geolocated textual information from Wikipedia articles. We demonstrate that modern NLP techniques can be used to predict community-level asset wealth and education outcomes using nearby geolocated Wikipedia articles. When paired with nightlights satellite imagery, our method outperforms all previously published benchmarks for this prediction task, indicating the potential of Wikipedia to inform both research in the social sciences and future policy decisions.

Hyperspectral imaging holds enormous potential to improve the state of the art in aerial vehicle tracking with low spatial and temporal resolutions. Recently, adaptive multimodal hyperspectral sensors have attracted growing interest due to their ability to record extended data quickly from aerial platforms. In this paper, we apply popular concepts from traditional object tracking, namely, kernelized correlation filters (KCFs) and deep convolutional neural network features to aerial tracking in the hyperspectral domain. We propose the deep hyperspectral KCF-based tracker (DeepHKCF) to efficiently track aerial vehicles using an adaptive multimodal hyperspectral sensor. We address low temporal resolution by designing a single KCF-in-multiple regions-of-interest (ROIs) approach to cover a reasonably large area. To increase the speed of deep convolutional features extraction from multiple ROIs, we design an effective ROI mapping strategy. The proposed tracker also provides flexibility to couple with the more advanced correlation filter trackers. The DeepHKCF tracker performs exceptionally well with deep features set up in a synthetic hyperspectral video generated by the digital imaging and remote sensing image generation (DIRSIG) software. In addition, we generate a large, synthetic, single-channel data set using DIRSIG to perform vehicle classification in the wide-area motion imagery (WAMI) platform. This way, the high fidelity of the DIRSIG software is proven, and a large-scale aerial vehicle classification data set is released to support studies on vehicle detection and tracking in the WAMI platform.

Computer vision technologies are very attractive for practical applications running on embedded systems. For such applications, it is desirable for the deployed algorithms to run in high-speed and require no offline training. To develop a single-target tracking algorithm with these properties, we propose an ensemble of the kernelized correlation filters (KCF), we call it EnKCF. A committee of KCFs is designed to specifically address the variations in scale and translation of moving objects. To guarantee a high-speed run-time performance, we deploy each of KCFs in turn, instead of applying multiple KCFs to each frame altogether. To reduce any potential drifts between individual KCFs' transition, we developed a particle filter. Experimental results showed that 1) the performance of ours is, on average, 70.10% for precision at 20 pixels, 53.00% for success rate for the OTB100 data, and 54.50% and 40.2% for the UAV123 data, and 2) our method is, on average, better than other high-speed trackers over 5% on precision on 20 pixels and 10-20% on AUC. In addition, our implementation ran at 340 fps for the OTB100 and at 416 fps for the UAV123 dataset that is faster than DCF (333 fps) for the OTB100 and KCF (296 fps) for the UAV123. To increase flexibility of the proposed EnKCF running on various platforms, we also explored different levels of deep convolutional features.