Data augmentation has been pivotal in successfully training deep learning models on classification tasks over the past decade. An important subclass of data augmentation techniques - which includes both label smoothing and Mixup - involves modifying not only the input data but also the input label during model training. In this work, we analyze the role played by the label augmentation aspect of such methods. We prove that linear models on linearly separable data trained with label augmentation learn only the minimum variance features in the data, while standard training (which includes weight decay) can learn higher variance features. An important consequence of our results is negative: label smoothing and Mixup can be less robust to adversarial perturbations of the training data when compared to standard training. We verify that our theory reflects practice via a range of experiments on synthetic data and image classification benchmarks.
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.
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.
Gradient inversion attacks aim to reconstruct local training data from intermediate gradients exposed in the federated learning framework. Despite successful attacks, all previous methods, starting from reconstructing a single data point and then relaxing the single-image limit to batch level, are only tested under hard label constraints. Even for single-image reconstruction, we still lack an analysis-based algorithm to recover augmented soft labels. In this work, we change the focus from enlarging batchsize to investigating the hard label constraints, considering a more realistic circumstance where label smoothing and mixup techniques are used in the training process. In particular, we are the first to initiate a novel algorithm to simultaneously recover the ground-truth augmented label and the input feature of the last fully-connected layer from single-input gradients, and provide a necessary condition for any analytical-based label recovery methods. Extensive experiments testify to the label recovery accuracy, as well as the benefits to the following image reconstruction. We believe soft labels in classification tasks are worth further attention in gradient inversion attacks.
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.
Vision language models (VLMs) have shown impressive capabilities across a variety of tasks, from logical reasoning to visual understanding. This opens the door to richer interaction with the world, for example robotic control. However, VLMs produce only textual outputs, while robotic control and other spatial tasks require outputting continuous coordinates, actions, or trajectories. How can we enable VLMs to handle such settings without fine-tuning on task-specific data? In this paper, we propose a novel visual prompting approach for VLMs that we call Prompting with Iterative Visual Optimization (PIVOT), which casts tasks as iterative visual question answering. In each iteration, the image is annotated with a visual representation of proposals that the VLM can refer to (e.g., candidate robot actions, localizations, or trajectories). The VLM then selects the best ones for the task. These proposals are iteratively refined, allowing the VLM to eventually zero in on the best available answer. We investigate PIVOT on real-world robotic navigation, real-world manipulation from images, instruction following in simulation, and additional spatial inference tasks such as localization. We find, perhaps surprisingly, that our approach enables zero-shot control of robotic systems without any robot training data, navigation in a variety of environments, and other capabilities. Although current performance is far from perfect, our work highlights potentials and limitations of this new regime and shows a promising approach for Internet-Scale VLMs in robotic and spatial reasoning domains. Website: pivot-prompt.github.io and HuggingFace: https://huggingface.co/spaces/pivot-prompt/pivot-prompt-demo.
Consider a multi-class labelling problem, where the labels can take values in $[k]$, and a predictor predicts a distribution over the labels. In this work, we study the following foundational question: Are there notions of multi-class calibration that give strong guarantees of meaningful predictions and can be achieved in time and sample complexities polynomial in $k$? Prior notions of calibration exhibit a tradeoff between computational efficiency and expressivity: they either suffer from having sample complexity exponential in $k$, or needing to solve computationally intractable problems, or give rather weak guarantees. Our main contribution is a notion of calibration that achieves all these desiderata: we formulate a robust notion of projected smooth calibration for multi-class predictions, and give new recalibration algorithms for efficiently calibrating predictors under this definition with complexity polynomial in $k$. Projected smooth calibration gives strong guarantees for all downstream decision makers who want to use the predictor for binary classification problems of the form: does the label belong to a subset $T \subseteq [k]$: e.g. is this an image of an animal? It ensures that the probabilities predicted by summing the probabilities assigned to labels in $T$ are close to some perfectly calibrated binary predictor for that task. We also show that natural strengthenings of our definition are computationally hard to achieve: they run into information theoretic barriers or computational intractability. Underlying both our upper and lower bounds is a tight connection that we prove between multi-class calibration and the well-studied problem of agnostic learning in the (standard) binary prediction setting.
Medical image segmentation takes an important position in various clinical applications. Deep learning has emerged as the predominant solution for automated segmentation of volumetric medical images. 2.5D-based segmentation models bridge computational efficiency of 2D-based models and spatial perception capabilities of 3D-based models. However, prevailing 2.5D-based models often treat each slice equally, failing to effectively learn and exploit inter-slice information, resulting in suboptimal segmentation performances. In this paper, a novel Momentum encoder-based inter-slice fusion transformer (MOSformer) is proposed to overcome this issue by leveraging inter-slice information at multi-scale feature maps extracted by different encoders. Specifically, dual encoders are employed to enhance feature distinguishability among different slices. One of the encoders is moving-averaged to maintain the consistency of slice representations. Moreover, an IF-Swin transformer module is developed to fuse inter-slice multi-scale features. The MOSformer is evaluated on three benchmark datasets (Synapse, ACDC, and AMOS), establishing a new state-of-the-art with 85.63%, 92.19%, and 85.43% of DSC, respectively. These promising results indicate its competitiveness in medical image segmentation. Codes and models of MOSformer will be made publicly available upon acceptance.
Generative domain adaptation has achieved remarkable progress, enabling us to adapt a pre-trained generator to a new target domain. However, existing methods simply adapt the generator to a single target domain and are limited to a single modality, either text-driven or image-driven. Moreover, they are prone to overfitting domain-specific attributes, which inevitably compromises cross-domain consistency. In this paper, we propose UniHDA, a unified and versatile framework for generative hybrid domain adaptation with multi-modal references from multiple domains. We use CLIP encoder to project multi-modal references into a unified embedding space and then linear interpolate the direction vectors from multiple target domains to achieve hybrid domain adaptation. To ensure the cross-domain consistency, we propose a novel cross-domain spatial structure (CSS) loss that maintains detailed spatial structure information between source and target generator. Experiments show that the adapted generator can synthesise realistic images with various attribute compositions. Additionally, our framework is versatile to multiple generators, \eg, StyleGAN2 and Diffusion Models.
This study introduces the 3D Residual-in-Residual Dense Block GAN (3D RRDB-GAN) for 3D super-resolution for radiology imagery. A key aspect of 3D RRDB-GAN is the integration of a 2.5D perceptual loss function, which contributes to improved volumetric image quality and realism. The effectiveness of our model was evaluated through 4x super-resolution experiments across diverse datasets, including Mice Brain MRH, OASIS, HCP1200, and MSD-Task-6. These evaluations, encompassing both quantitative metrics like LPIPS and FID and qualitative assessments through sample visualizations, demonstrate the models effectiveness in detailed image analysis. The 3D RRDB-GAN offers a significant contribution to medical imaging, particularly by enriching the depth, clarity, and volumetric detail of medical images. Its application shows promise in enhancing the interpretation and analysis of complex medical imagery from a comprehensive 3D perspective.