In light of recent advances in multimodal Large Language Models (LLMs), there is increasing attention to scaling them from image-text data to more informative real-world videos. Compared to static images, video poses unique challenges for effective large-scale pre-training due to the modeling of its spatiotemporal dynamics. In this paper, we address such limitations in video-language pre-training with an efficient video decomposition that represents each video as keyframes and temporal motions. These are then adapted to an LLM using well-designed tokenizers that discretize visual and temporal information as a few tokens, thus enabling unified generative pre-training of videos, images, and text. At inference, the generated tokens from the LLM are carefully recovered to the original continuous pixel space to create various video content. Our proposed framework is both capable of comprehending and generating image and video content, as demonstrated by its competitive performance across 13 multimodal benchmarks in image and video understanding and generation. Our code and models will be available at https://video-lavit.github.io.
Can a lightweight Vision Transformer (ViT) match or exceed the performance of Convolutional Neural Networks (CNNs) like ResNet on small datasets with small image resolutions? This report demonstrates that a pure ViT can indeed achieve superior performance through pre-training, using a masked auto-encoder technique with minimal image scaling. Our experiments on the CIFAR-10 and CIFAR-100 datasets involved ViT models with fewer than 3.65 million parameters and a multiply-accumulate (MAC) count below 0.27G, qualifying them as 'lightweight' models. Unlike previous approaches, our method attains state-of-the-art performance among similar lightweight transformer-based architectures without significantly scaling up images from CIFAR-10 and CIFAR-100. This achievement underscores the efficiency of our model, not only in handling small datasets but also in effectively processing images close to their original scale.
Inspired by the outstanding zero-shot capability of vision language models (VLMs) in image classification tasks, open-vocabulary object detection has attracted increasing interest by distilling the broad VLM knowledge into detector training. However, most existing open-vocabulary detectors learn by aligning region embeddings with categorical labels (e.g., bicycle) only, disregarding the capability of VLMs on aligning visual embeddings with fine-grained text description of object parts (e.g., pedals and bells). This paper presents DVDet, a Descriptor-Enhanced Open Vocabulary Detector that introduces conditional context prompts and hierarchical textual descriptors that enable precise region-text alignment as well as open-vocabulary detection training in general. Specifically, the conditional context prompt transforms regional embeddings into image-like representations that can be directly integrated into general open vocabulary detection training. In addition, we introduce large language models as an interactive and implicit knowledge repository which enables iterative mining and refining visually oriented textual descriptors for precise region-text alignment. Extensive experiments over multiple large-scale benchmarks show that DVDet outperforms the state-of-the-art consistently by large margins.
In recent years, neural networks have been used to solve phase retrieval problems in imaging with superior accuracy and speed than traditional techniques, especially in the presence of noise. However, in the context of interferometric imaging, phase noise has been largely unaddressed by existing neural network architectures. Such noise arises naturally in an interferometer due to mechanical instabilities or atmospheric turbulence, limiting measurement acquisition times and posing a challenge in scenarios with limited light intensity, such as remote sensing. Here, we introduce a 3D-2D Phase Retrieval U-Net (PRUNe) that takes noisy and randomly phase-shifted interferograms as inputs, and outputs a single 2D phase image. A 3D downsampling convolutional encoder captures correlations within and between frames to produce a 2D latent space, which is upsampled by a 2D decoder into a phase image. We test our model against a state-of-the-art singular value decomposition algorithm and find PRUNe reconstructions consistently show more accurate and smooth reconstructions, with a x2.5 - 4 lower mean squared error at multiple signal-to-noise ratios for interferograms with low (< 1 photon/pixel) and high (~100 photons/pixel) signal intensity. Our model presents a faster and more accurate approach to perform phase retrieval in extremely low light intensity interferometry in presence of phase noise, and will find application in other multi-frame noisy imaging techniques.
Vision-Language Models (VLMs) have emerged as the state-of-the-art representation learning solution, with myriads of downstream applications such as image classification, retrieval and generation. A natural question is whether these models memorize their training data, which also has implications for generalization. We propose a new method for measuring memorization in VLMs, which we call d\'ej\`a vu memorization. For VLMs trained on image-caption pairs, we show that the model indeed retains information about individual objects in the training images beyond what can be inferred from correlations or the image caption. We evaluate d\'ej\`a vu memorization at both sample and population level, and show that it is significant for OpenCLIP trained on as many as 50M image-caption pairs. Finally, we show that text randomization considerably mitigates memorization while only moderately impacting the model's downstream task performance.
The usage of Artificial Intelligence (AI) systems has increased exponentially, thanks to their ability to reduce the amount of data to be analyzed, the user efforts and preserving a high rate of accuracy. However, introducing this new element in the loop has converted them into attacked points that can compromise the reliability of the systems. This new scenario has raised crucial challenges regarding the reliability and trustworthiness of the AI models, as well as about the uncertainties in their response decisions, becoming even more crucial when applied in critical domains such as healthcare, chemical, electrical plants, etc. To contain these issues, in this paper, we present NeuralSentinel (NS), a tool able to validate the reliability and trustworthiness of AI models. This tool combines attack and defence strategies and explainability concepts to stress an AI model and help non-expert staff increase their confidence in this new system by understanding the model decisions. NS provide a simple and easy-to-use interface for helping humans in the loop dealing with all the needed information. This tool was deployed and used in a Hackathon event to evaluate the reliability of a skin cancer image detector. During the event, experts and non-experts attacked and defended the detector, learning which factors were the most important for model misclassification and which techniques were the most efficient. The event was also used to detect NS's limitations and gather feedback for further improvements.
The histopathology analysis is of great significance for the diagnosis and prognosis of cancers, however, it has great challenges due to the enormous heterogeneity of gigapixel whole slide images (WSIs) and the intricate representation of pathological features. However, recent methods have not adequately exploited geometrical representation in WSIs which is significant in disease diagnosis. Therefore, we proposed a novel weakly-supervised framework, Geometry-Aware Transformer (GOAT), in which we urge the model to pay attention to the geometric characteristics within the tumor microenvironment which often serve as potent indicators. In addition, a context-aware attention mechanism is designed to extract and enhance the morphological features within WSIs.
This article studies the problem of image segmentation-based semantic communication in autonomous driving. In real traffic scenes, detecting the key objects (e.g., vehicles, pedestrians and obstacles) is more crucial than that of other objects to guarantee driving safety. Therefore, we propose a vehicular image segmentation-oriented semantic communication system, termed VIS-SemCom, where image segmentation features of important objects are transmitted to reduce transmission redundancy. First, to accurately extract image semantics, we develop a semantic codec based on Swin Transformer architecture, which expands the perceptual field thus improving the segmentation accuracy. Next, we propose a multi-scale semantic extraction scheme via assigning the number of Swin Transformer blocks for diverse resolution features, thus highlighting the important objects' accuracy. Furthermore, the importance-aware loss is invoked to emphasize the important objects, and an online hard sample mining (OHEM) strategy is proposed to handle small sample issues in the dataset. Experimental results demonstrate that the proposed VIS-SemCom can achieve a coding gain of nearly 6 dB with a 60% mean intersection over union (mIoU), reduce the transmitted data amount by up to 70% with a 60% mIoU, and improve the segmentation intersection over union (IoU) of important objects by 4%, compared to traditional transmission scheme.
Tumor segmentation from multi-modal brain MRI images is a challenging task due to the limited samples, high variance in shapes and uneven distribution of tumor morphology. The performance of automated medical image segmentation has been significant improvement by the recent advances in deep learning. However, the model predictions have not yet reached the desired level for clinical use in terms of accuracy and generalizability. In order to address the distinct problems presented in Challenges 1, 2, and 3 of BraTS 2023, we have constructed an optimization framework based on a 3D U-Net model for brain tumor segmentation. This framework incorporates a range of techniques, including various pre-processing and post-processing techniques, and transfer learning. On the validation datasets, this multi-modality brain tumor segmentation framework achieves an average lesion-wise Dice score of 0.79, 0.72, 0.74 on Challenges 1, 2, 3 respectively.
Despite recent improvements in audio-text modeling, audio-text contrastive models still lag behind their image-text counterparts in scale and performance. We propose a method to improve both the scale and the training of audio-text contrastive models. Specifically, we craft a large-scale audio-text dataset consisting of over 13,000 hours of text-labeled audio, aided by large language model (LLM) processing and audio captioning. Further, we employ an masked autoencoder (MAE) pre-pretraining phase with random patch dropout, which allows us to both scale unlabeled audio datasets and train efficiently with variable length audio. After MAE pre-pretraining of our audio encoder, we train a contrastive model with an auxiliary captioning objective. Our final model, which we name Cacophony, achieves state-of-the-art performance on audio-text retrieval tasks, and exhibits competitive results on other downstream tasks such as zero-shot classification.