This work introduces CAPIVARA, a cost-efficient framework designed to enhance the performance of multilingual CLIP models in low-resource languages. While CLIP has excelled in zero-shot vision-language tasks, the resource-intensive nature of model training remains challenging. Many datasets lack linguistic diversity, featuring solely English descriptions for images. CAPIVARA addresses this by augmenting text data using image captioning and machine translation to generate multiple synthetic captions in low-resource languages. We optimize the training pipeline with LiT, LoRA, and gradient checkpointing to alleviate the computational cost. Through extensive experiments, CAPIVARA emerges as state of the art in zero-shot tasks involving images and Portuguese texts. We show the potential for significant improvements in other low-resource languages, achieved by fine-tuning the pre-trained multilingual CLIP using CAPIVARA on a single GPU for 2 hours. Our model and code is available at https://github.com/hiaac-nlp/CAPIVARA.
We present Contextualized Local Visual Embeddings (CLoVE), a self-supervised convolutional-based method that learns representations suited for dense prediction tasks. CLoVE deviates from current methods and optimizes a single loss function that operates at the level of contextualized local embeddings learned from output feature maps of convolution neural network (CNN) encoders. To learn contextualized embeddings, CLoVE proposes a normalized mult-head self-attention layer that combines local features from different parts of an image based on similarity. We extensively benchmark CLoVE's pre-trained representations on multiple datasets. CLoVE reaches state-of-the-art performance for CNN-based architectures in 4 dense prediction downstream tasks, including object detection, instance segmentation, keypoint detection, and dense pose estimation.
The intersection of vision and language is of major interest due to the increased focus on seamless integration between recognition and reasoning. Scene graphs (SGs) have emerged as a useful tool for multimodal image analysis, showing impressive performance in tasks such as Visual Question Answering (VQA). In this work, we demonstrate that despite the effectiveness of scene graphs in VQA tasks, current methods that utilize idealized annotated scene graphs struggle to generalize when using predicted scene graphs extracted from images. To address this issue, we introduce the SelfGraphVQA framework. Our approach extracts a scene graph from an input image using a pre-trained scene graph generator and employs semantically-preserving augmentation with self-supervised techniques. This method improves the utilization of graph representations in VQA tasks by circumventing the need for costly and potentially biased annotated data. By creating alternative views of the extracted graphs through image augmentations, we can learn joint embeddings by optimizing the informational content in their representations using an un-normalized contrastive approach. As we work with SGs, we experiment with three distinct maximization strategies: node-wise, graph-wise, and permutation-equivariant regularization. We empirically showcase the effectiveness of the extracted scene graph for VQA and demonstrate that these approaches enhance overall performance by highlighting the significance of visual information. This offers a more practical solution for VQA tasks that rely on SGs for complex reasoning questions.
To mitigate the necessity for large amounts of supervised segmentation annotation sets, multiple Weakly Supervised Semantic Segmentation (WSSS) strategies have been devised. These will often rely on advanced data and model regularization strategies to instigate the development of useful properties (e.g., prediction completeness and fidelity to semantic boundaries) in segmentation priors, notwithstanding the lack of annotated information. In this work, we first create a strong baseline by analyzing complementary WSSS techniques and regularizing strategies, considering their strengths and limitations. We then propose a new Class-specific Adversarial Erasing strategy, comprising two adversarial CAM generating networks being gradually refined to produce robust semantic segmentation proposals. Empirical results suggest that our approach induces substantial improvement in the effectiveness of the baseline, resulting in a noticeable improvement over both Pascal VOC 2012 and MS COCO 2014 datasets.
Early and accurate diagnosis of COVID-19 is essential to control the rapid spread of the pandemic and mitigate sequelae in the population. Current diagnostic methods, such as RT-PCR, are effective but require time to provide results and can quickly overwhelm clinics, requiring individual laboratory analysis. Automatic detection methods have the potential to significantly reduce diagnostic time. To this end, learning-based methods using lung imaging have been explored. Although they require specialized hardware, automatic evaluation methods can be performed simultaneously, making diagnosis faster. Convolutional neural networks have been widely used to detect pneumonia caused by COVID-19 in lung images. This work describes an architecture based on 3D convolutional neural networks for detecting COVID-19 in computed tomography images. Despite the challenging scenario present in the dataset, the results obtained with our architecture demonstrated to be quite promising.
Pruning is a standard technique for reducing the computational cost of deep networks. Many advances in pruning leverage concepts from the Lottery Ticket Hypothesis (LTH). LTH reveals that inside a trained dense network exists sparse subnetworks (tickets) able to achieve similar accuracy (i.e., win the lottery - winning tickets). Pruning at initialization focuses on finding winning tickets without training a dense network. Studies on these concepts share the trend that subnetworks come from weight or filter pruning. In this work, we investigate LTH and pruning at initialization from the lens of layer pruning. First, we confirm the existence of winning tickets when the pruning process removes layers. Leveraged by this observation, we propose to discover these winning tickets at initialization, eliminating the requirement of heavy computational resources for training the initial (over-parameterized) dense network. Extensive experiments show that our winning tickets notably speed up the training phase and reduce up to 51% of carbon emission, an important step towards democratization and green Artificial Intelligence. Beyond computational benefits, our winning tickets exhibit robustness against adversarial and out-of-distribution examples. Finally, we show that our subnetworks easily win the lottery at initialization while tickets from filter removal (the standard structured LTH) hardly become winning tickets.
Single image super-resolution (SISR) is the process of obtaining one high-resolution version of a low-resolution image by increasing the number of pixels per unit area. This method has been actively investigated by the research community, due to the wide variety of real-world problems where it can be applied, from aerial and satellite imaging to compressed image and video enhancement. Despite the improvements achieved by deep learning in the field, the vast majority of the used networks are based on traditional convolutions, with the solutions focusing on going deeper and/or wider, and innovations coming from jointly employing successful concepts from other fields. In this work, we decided to step up from the traditional convolutions and adopt the concept of capsules. Since their overwhelming results both in image classification and segmentation problems, we question how suitable they are for SISR. We also verify that different solutions share most of their configurations, and argue that this trend leads to fewer explorations of network varieties. During our experiments, we check various strategies to improve results, ranging from new and different loss functions to changes in the capsule layers. Our network achieved good results with fewer convolutional-based layers, showing that capsules might be a concept worth applying in the image super-resolution problem.
The success of Zero-shot Action Recognition (ZSAR) methods is intrinsically related to the nature of semantic side information used to transfer knowledge, although this aspect has not been primarily investigated in the literature. This work introduces a new ZSAR method based on the relationships of actions-objects and actions-descriptive sentences. We demonstrate that representing all object classes using descriptive sentences generates an accurate object-action affinity estimation when a paraphrase estimation method is used as an embedder. We also show how to estimate probabilities over the set of action classes based only on a set of sentences without hard human labeling. In our method, the probabilities from these two global classifiers (i.e., which use features computed over the entire video) are combined, producing an efficient transfer knowledge model for action classification. Our results are state-of-the-art in the Kinetics-400 dataset and are competitive on UCF-101 under the ZSAR evaluation. Our code is available at https://github.com/valterlej/objsentzsar
RGB-D data is essential for solving many problems in computer vision. Hundreds of public RGB-D datasets containing various scenes, such as indoor, outdoor, aerial, driving, and medical, have been proposed. These datasets are useful for different applications and are fundamental for addressing classic computer vision tasks, such as monocular depth estimation. This paper reviewed and categorized image datasets that include depth information. We gathered 203 datasets that contain accessible data and grouped them into three categories: scene/objects, body, and medical. We also provided an overview of the different types of sensors, depth applications, and we examined trends and future directions of the usage and creation of datasets containing depth data, and how they can be applied to investigate the development of generalizable machine learning models in the monocular depth estimation field.
Recently, several approaches have explored the detection and classification of objects in videos to perform Zero-Shot Action Recognition with remarkable results. In these methods, class-object relationships are used to associate visual patterns with the semantic side information because these relationships also tend to appear in texts. Therefore, word vector methods would reflect them in their latent representations. Inspired by these methods and by video captioning's ability to describe events not only with a set of objects but with contextual information, we propose a method in which video captioning models, called observers, provide different and complementary descriptive sentences. We demonstrate that representing videos with descriptive sentences instead of deep features, in ZSAR, is viable and naturally alleviates the domain adaptation problem, as we reached state-of-the-art (SOTA) performance on the UCF101 dataset and competitive performance on HMDB51 without their training sets. We also demonstrate that word vectors are unsuitable for building the semantic embedding space of our descriptions. Thus, we propose to represent the classes with sentences extracted from documents acquired with search engines on the Internet, without any human evaluation on the quality of descriptions. Lastly, we build a shared semantic space employing BERT-based embedders pre-trained in the paraphrasing task on multiple text datasets. We show that this pre-training is essential for bridging the semantic gap. The projection onto this space is straightforward for both types of information, visual and semantic, because they are sentences, enabling the classification with nearest neighbour rule in this shared space. Our code is available at https://github.com/valterlej/zsarcap.