Normalization techniques have been widely used in the field of deep learning due to their capability of enabling higher learning rates and are less careful in initialization. However, the effectiveness of popular normalization technologies is typically limited to specific areas. Unlike the standard Batch Normalization (BN) and Layer Normalization (LN), where BN computes the mean and variance along the (N,H,W) dimensions and LN computes the mean and variance along the (C,H,W) dimensions (N, C, H and W are the batch, channel, spatial height and width dimension, respectively), this paper presents a novel normalization technique called Batch Channel Normalization (BCN). To exploit both the channel and batch dependence and adaptively and combine the advantages of BN and LN based on specific datasets or tasks, BCN separately normalizes inputs along the (N, H, W) and (C, H, W) axes, then combines the normalized outputs based on adaptive parameters. As a basic block, BCN can be easily integrated into existing models for various applications in the field of computer vision. Empirical results show that the proposed technique can be seamlessly applied to various versions of CNN or Vision Transformer architecture. The code is publicly available at https://github.com/AfifaKhaled/BatchChannel-Normalization
Catastrophic forgetting is a significant challenge in the field of machine learning, particularly in neural networks. When a neural network learns to perform well on a new task, it often forgets its previously acquired knowledge or experiences. This phenomenon occurs because the network adjusts its weights and connections to minimize the loss on the new task, which can inadvertently overwrite or disrupt the representations that were crucial for the previous tasks. As a result, the the performance of the network on earlier tasks deteriorates, limiting its ability to learn and adapt to a sequence of tasks. In this paper, we propose a novel method for preventing catastrophic forgetting in machine learning applications, specifically focusing on neural networks. Our approach aims to preserve the knowledge of the network across multiple tasks while still allowing it to learn new information effectively. We demonstrate the effectiveness of our method by conducting experiments on various benchmark datasets, including Split MNIST, Split CIFAR10, Split Fashion MNIST, and Split CIFAR100. These datasets are created by dividing the original datasets into separate, non overlapping tasks, simulating a continual learning scenario where the model needs to learn multiple tasks sequentially without forgetting the previous ones. Our proposed method tackles the catastrophic forgetting problem by incorporating negotiated representations into the learning process, which allows the model to maintain a balance between retaining past experiences and adapting to new tasks. By evaluating our method on these challenging datasets, we aim to showcase its potential for addressing catastrophic forgetting and improving the performance of neural networks in continual learning settings.
The accelerated advancement of generative AI significantly enhance the viability and effectiveness of generative regional editing methods. This evolution render the image manipulation more accessible, thereby intensifying the risk of altering the conveyed information within original images and even propagating misinformation. Consequently, there exists a critical demand for robust capable of detecting the edited images. However, the lack of comprehensive dataset containing images edited with abundant and advanced generative regional editing methods poses a substantial obstacle to the advancement of corresponding detection methods. We endeavor to fill the vacancy by constructing the GRE dataset, a large-scale generative regional editing dataset with the following advantages: 1) Collection of real-world original images, focusing on two frequently edited scenarios. 2) Integration of a logical and simulated editing pipeline, leveraging multiple large models in various modalities. 3) Inclusion of various editing approaches with distinct architectures. 4) Provision of comprehensive analysis tasks. We perform comprehensive experiments with proposed three tasks: edited image classification, edited method attribution and edited region localization, providing analysis of distinct editing methods and evaluation of detection methods in related fields. We expect that the GRE dataset can promote further research and exploration in the field of generative region editing detection.
Fine-tuning vision-language models (VLMs) like CLIP to downstream tasks is often necessary to optimize their performance. However, a major obstacle is the limited availability of labeled data. We study the use of pseudolabels, i.e., heuristic labels for unlabeled data, to enhance CLIP via prompt tuning. Conventional pseudolabeling trains a model on labeled data and then generates labels for unlabeled data. VLMs' zero-shot capabilities enable a "second generation" of pseudolabeling approaches that do not require task-specific training on labeled data. By using zero-shot pseudolabels as a source of supervision, we observe that learning paradigms such as semi-supervised, transductive zero-shot, and unsupervised learning can all be seen as optimizing the same loss function. This unified view enables the development of versatile training strategies that are applicable across learning paradigms. We investigate them on image classification tasks where CLIP exhibits limitations, by varying prompt modalities, e.g., textual or visual prompts, and learning paradigms. We find that (1) unexplored prompt tuning strategies that iteratively refine pseudolabels consistently improve CLIP accuracy, by 19.5 points in semi-supervised learning, by 28.4 points in transductive zero-shot learning, and by 15.2 points in unsupervised learning, and (2) unlike conventional semi-supervised pseudolabeling, which exacerbates model biases toward classes with higher-quality pseudolabels, prompt tuning leads to a more equitable distribution of per-class accuracy. The code to reproduce the experiments is at https://github.com/BatsResearch/menghini-neurips23-code.
Recent progress in scaling up large language models has shown impressive capabilities in performing few-shot learning across a wide range of text-based tasks. However, a key limitation is that these language models fundamentally lack visual perception - a crucial attribute needed to extend these models to be able to interact with the real world and solve vision tasks, such as in visual-question answering and robotics. Prior works have largely connected image to text through pretraining and/or fine-tuning on curated image-text datasets, which can be a costly and expensive process. In order to resolve this limitation, we propose a simple yet effective approach called Language-Quantized AutoEncoder (LQAE), a modification of VQ-VAE that learns to align text-image data in an unsupervised manner by leveraging pretrained language models (e.g., BERT, RoBERTa). Our main idea is to encode image as sequences of text tokens by directly quantizing image embeddings using a pretrained language codebook. We then apply random masking followed by a BERT model, and have the decoder reconstruct the original image from BERT predicted text token embeddings. By doing so, LQAE learns to represent similar images with similar clusters of text tokens, thereby aligning these two modalities without the use of aligned text-image pairs. This enables few-shot image classification with large language models (e.g., GPT-3) as well as linear classification of images based on BERT text features. To the best of our knowledge, our work is the first work that uses unaligned images for multimodal tasks by leveraging the power of pretrained language models.
Irregularly sampled time series are increasingly prevalent, particularly in medical domains. While various specialized methods have been developed to handle these irregularities, effectively modeling their complex dynamics and pronounced sparsity remains a challenge. This paper introduces a novel perspective by converting irregularly sampled time series into line graph images, then utilizing powerful pre-trained vision transformers for time series classification in the same way as image classification. This method not only largely simplifies specialized algorithm designs but also presents the potential to serve as a universal framework for time series modeling. Remarkably, despite its simplicity, our approach outperforms state-of-the-art specialized algorithms on several popular healthcare and human activity datasets. Especially in the rigorous leave-sensors-out setting where a portion of variables is omitted during testing, our method exhibits strong robustness against varying degrees of missing observations, achieving an impressive improvement of 42.8% in absolute F1 score points over leading specialized baselines even with half the variables masked. Code and data are available at https://github.com/Leezekun/ViTST
Despite the growing use of transformer models in computer vision, a mechanistic understanding of these networks is still needed. This work introduces a method to reverse-engineer Vision Transformers trained to solve image classification tasks. Inspired by previous research in NLP, we demonstrate how the inner representations at any level of the hierarchy can be projected onto the learned class embedding space to uncover how these networks build categorical representations for their predictions. We use our framework to show how image tokens develop class-specific representations that depend on attention mechanisms and contextual information, and give insights on how self-attention and MLP layers differentially contribute to this categorical composition. We additionally demonstrate that this method (1) can be used to determine the parts of an image that would be important for detecting the class of interest, and (2) exhibits significant advantages over traditional linear probing approaches. Taken together, our results position our proposed framework as a powerful tool for mechanistic interpretability and explainability research.
Due to the high price and heavy energy consumption of GPUs, deploying deep models on IoT devices such as microcontrollers makes significant contributions for ecological AI. Conventional methods successfully enable convolutional neural network inference of high resolution images on microcontrollers, while the framework for vision transformers that achieve the state-of-the-art performance in many vision applications still remains unexplored. In this paper, we propose a hardware-algorithm co-optimizations method called MCUFormer to deploy vision transformers on microcontrollers with extremely limited memory, where we jointly design transformer architecture and construct the inference operator library to fit the memory resource constraint. More specifically, we generalize the one-shot network architecture search (NAS) to discover the optimal architecture with highest task performance given the memory budget from the microcontrollers, where we enlarge the existing search space of vision transformers by considering the low-rank decomposition dimensions and patch resolution for memory reduction. For the construction of the inference operator library of vision transformers, we schedule the memory buffer during inference through operator integration, patch embedding decomposition, and token overwriting, allowing the memory buffer to be fully utilized to adapt to the forward pass of the vision transformer. Experimental results demonstrate that our MCUFormer achieves 73.62\% top-1 accuracy on ImageNet for image classification with 320KB memory on STM32F746 microcontroller. Code is available at https://github.com/liangyn22/MCUFormer.
The advent of high-capacity pre-trained models has revolutionized problem-solving in computer vision, shifting the focus from training task-specific models to adapting pre-trained models. Consequently, effectively adapting large pre-trained models to downstream tasks in an efficient manner has become a prominent research area. Existing solutions primarily concentrate on designing lightweight adapters and their interaction with pre-trained models, with the goal of minimizing the number of parameters requiring updates. In this study, we propose a novel Adapter Re-Composing (ARC) strategy that addresses efficient pre-trained model adaptation from a fresh perspective. Our approach considers the reusability of adaptation parameters and introduces a parameter-sharing scheme. Specifically, we leverage symmetric down-/up-projections to construct bottleneck operations, which are shared across layers. By learning low-dimensional re-scaling coefficients, we can effectively re-compose layer-adaptive adapters. This parameter-sharing strategy in adapter design allows us to significantly reduce the number of new parameters while maintaining satisfactory performance, thereby offering a promising approach to compress the adaptation cost. We conduct experiments on 24 downstream image classification tasks using various Vision Transformer variants to evaluate our method. The results demonstrate that our approach achieves compelling transfer learning performance with a reduced parameter count. Our code is available at \href{https://github.com/DavidYanAnDe/ARC}{https://github.com/DavidYanAnDe/ARC}.
Deeper Vision Transformers (ViTs) are more challenging to train. We expose a degradation problem in deeper layers of ViT when using masked image modeling (MIM) for pre-training. To ease the training of deeper ViTs, we introduce a self-supervised learning framework called Masked Image Residual Learning (MIRL), which significantly alleviates the degradation problem, making scaling ViT along depth a promising direction for performance upgrade. We reformulate the pre-training objective for deeper layers of ViT as learning to recover the residual of the masked image. We provide extensive empirical evidence showing that deeper ViTs can be effectively optimized using MIRL and easily gain accuracy from increased depth. With the same level of computational complexity as ViT-Base and ViT-Large, we instantiate 4.5$\times$ and 2$\times$ deeper ViTs, dubbed ViT-S-54 and ViT-B-48. The deeper ViT-S-54, costing 3$\times$ less than ViT-Large, achieves performance on par with ViT-Large. ViT-B-48 achieves 86.2% top-1 accuracy on ImageNet. On one hand, deeper ViTs pre-trained with MIRL exhibit excellent generalization capabilities on downstream tasks, such as object detection and semantic segmentation. On the other hand, MIRL demonstrates high pre-training efficiency. With less pre-training time, MIRL yields competitive performance compared to other approaches.
Masked image modeling (MIM) has become a prevalent pre-training setup for vision foundation models and attains promising performance. Despite its success, existing MIM methods discard the decoder network during downstream applications, resulting in inconsistent representations between pre-training and fine-tuning and can hamper downstream task performance. In this paper, we propose a new architecture, RevColV2, which tackles this issue by keeping the entire autoencoder architecture during both pre-training and fine-tuning. The main body of RevColV2 contains bottom-up columns and top-down columns, between which information is reversibly propagated and gradually disentangled. Such design enables our architecture with the nice property: maintaining disentangled low-level and semantic information at the end of the network in MIM pre-training. Our experimental results suggest that a foundation model with decoupled features can achieve competitive performance across multiple downstream vision tasks such as image classification, semantic segmentation and object detection. For example, after intermediate fine-tuning on ImageNet-22K dataset, RevColV2-L attains 88.4% top-1 accuracy on ImageNet-1K classification and 58.6 mIoU on ADE20K semantic segmentation. With extra teacher and large scale dataset, RevColv2-L achieves 62.1 box AP on COCO detection and 60.4 mIoU on ADE20K semantic segmentation. Code and models are released at https://github.com/megvii-research/RevCol
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.
Scaling laws have been recently employed to derive compute-optimal model size (number of parameters) for a given compute duration. We advance and refine such methods to infer compute-optimal model shapes, such as width and depth, and successfully implement this in vision transformers. Our shape-optimized vision transformer, SoViT, achieves results competitive with models that exceed twice its size, despite being pre-trained with an equivalent amount of compute. For example, SoViT-400m/14 achieves 90.3% fine-tuning accuracy on ILSRCV2012, surpassing the much larger ViT-g/14 and approaching ViT-G/14 under identical settings, with also less than half the inference cost. We conduct a thorough evaluation across multiple tasks, such as image classification, captioning, VQA and zero-shot transfer, demonstrating the effectiveness of our model across a broad range of domains and identifying limitations. Overall, our findings challenge the prevailing approach of blindly scaling up vision models and pave a path for a more informed scaling.
Task arithmetic has recently emerged as a cost-effective and scalable approach to edit pre-trained models directly in weight space: By adding the fine-tuned weights of different tasks, the model's performance can be improved on these tasks, while negating them leads to task forgetting. Yet, our understanding of the effectiveness of task arithmetic and its underlying principles remains limited. We present a comprehensive study of task arithmetic in vision-language models and show that weight disentanglement is the crucial factor that makes it effective. This property arises during pre-training and manifests when distinct directions in weight space govern separate, localized regions in function space associated with the tasks. Notably, we show that fine-tuning models in their tangent space by linearizing them amplifies weight disentanglement. This leads to substantial performance improvements across multiple task arithmetic benchmarks and diverse models. Building on these findings, we provide theoretical and empirical analyses of the neural tangent kernel (NTK) of these models and establish a compelling link between task arithmetic and the spatial localization of the NTK eigenfunctions. Overall, our work uncovers novel insights into the fundamental mechanisms of task arithmetic and offers a more reliable and effective approach to edit pre-trained models through the NTK linearization.
Neural architecture search has proven to be a powerful approach to designing and refining neural networks, often boosting their performance and efficiency over manually-designed variations, but comes with computational overhead. While there has been a considerable amount of research focused on lowering the cost of NAS for mainstream tasks, such as image classification, a lot of those improvements stem from the fact that those tasks are well-studied in the broader context. Consequently, applicability of NAS to emerging and under-represented domains is still associated with a relatively high cost and/or uncertainty about the achievable gains. To address this issue, we turn our focus towards the recent growth of publicly available NAS benchmarks in an attempt to extract general NAS knowledge, transferable across different tasks and search spaces. We borrow from the rich field of meta-learning for few-shot adaptation and carefully study applicability of those methods to NAS, with a special focus on the relationship between task-level correlation (domain shift) and predictor transferability; which we deem critical for improving NAS on diverse tasks. In our experiments, we use 6 NAS benchmarks in conjunction, spanning in total 16 NAS settings -- our meta-learning approach not only shows superior (or matching) performance in the cross-validation experiments but also successful extrapolation to a new search space and tasks.
Collaborative inference systems are one of the emerging solutions for deploying deep neural networks (DNNs) at the wireless network edge. Their main idea is to divide a DNN into two parts, where the first is shallow enough to be reliably executed at edge devices of limited computational power, while the second part is executed at an edge server with higher computational capabilities. The main advantage of such systems is that the input of the DNN gets compressed as the subsequent layers of the shallow part extract only the information necessary for the task. As a result, significant communication savings can be achieved compared to transmitting raw input samples. In this work, we study early exiting in the context of collaborative inference, which allows obtaining inference results at the edge device for certain samples, without the need to transmit the partially processed data to the edge server at all, leading to further communication savings. The central part of our system is the transmission-decision (TD) mechanism, which, given the information from the early exit, and the wireless channel conditions, decides whether to keep the early exit prediction or transmit the data to the edge server for further processing. In this paper, we evaluate various TD mechanisms and show experimentally, that for an image classification task over the wireless edge, proper utilization of early exits can provide both performance gains and significant communication savings.
Instruction tuning large language models (LLMs) using machine-generated instruction-following data has improved zero-shot capabilities on new tasks, but the idea is less explored in the multimodal field. In this paper, we present the first attempt to use language-only GPT-4 to generate multimodal language-image instruction-following data. By instruction tuning on such generated data, we introduce LLaVA: Large Language and Vision Assistant, an end-to-end trained large multimodal model that connects a vision encoder and LLM for general-purpose visual and language understanding.Our early experiments show that LLaVA demonstrates impressive multimodel chat abilities, sometimes exhibiting the behaviors of multimodal GPT-4 on unseen images/instructions, and yields a 85.1% relative score compared with GPT-4 on a synthetic multimodal instruction-following dataset. When fine-tuned on Science QA, the synergy of LLaVA and GPT-4 achieves a new state-of-the-art accuracy of 92.53%. We make GPT-4 generated visual instruction tuning data, our model and code base publicly available.
Graph neural networks (GNNs) present a promising alternative to CNNs and transformers in certain image processing applications due to their parameter-efficiency in modeling spatial relationships. Currently, a major area of research involves the converting non-graph input data for GNN-based models, notably in scenarios where the data originates from images. One approach involves converting images into nodes by identifying significant keypoints within them. Super-Retina, a semi-supervised technique, has been utilized for detecting keypoints in retinal images. However, its limitations lie in the dependency on a small initial set of ground truth keypoints, which is progressively expanded to detect more keypoints. Having encountered difficulties in detecting consistent initial keypoints in brain images using SIFT and LoFTR, we proposed a new approach: radiomic feature-based keypoint detection. Demonstrating the anatomical significance of the detected keypoints was achieved by showcasing their efficacy in improving registration processes guided by these keypoints. Subsequently, these keypoints were employed as the ground truth for the keypoint detection method (LK-SuperRetina). Furthermore, the study showcases the application of GNNs in image matching, highlighting their superior performance in terms of both the number of good matches and confidence scores. This research sets the stage for expanding GNN applications into various other applications, including but not limited to image classification, segmentation, and registration.
Continual test-time adaptation (cTTA) methods are designed to facilitate the continual adaptation of models to dynamically changing real-world environments where computational resources are limited. Due to this inherent limitation, existing approaches fail to simultaneously achieve accuracy and efficiency. In detail, when using a single image, the instability caused by batch normalization layers and entropy loss significantly destabilizes many existing methods in real-world cTTA scenarios. To overcome these challenges, we present BESTTA, a novel single image continual test-time adaptation method guided by style transfer, which enables stable and efficient adaptation to the target environment by transferring the style of the input image to the source style. To implement the proposed method, we devise BeIN, a simple yet powerful normalization method, along with the style-guided losses. We demonstrate that BESTTA effectively adapts to the continually changing target environment, leveraging only a single image on both semantic segmentation and image classification tasks. Remarkably, despite training only two parameters in a BeIN layer consuming the least memory, BESTTA outperforms existing state-of-the-art methods in terms of performance.
In many practical computer vision scenarios unlabeled data is plentiful, but labels are scarce and difficult to obtain. As a result, semi-supervised learning which leverages unlabeled data to boost the performance of supervised classifiers have received significant attention in recent literature. One major class of semi-supervised algorithms is co-training. In co-training two different models leverage different independent and sufficient "views" of the data to jointly make better predictions. During co-training each model creates pseudo labels on unlabeled points which are used to improve the other model. We show that in the common case when independent views are not available we can construct such views inexpensively using pre-trained models. Co-training on the constructed views yields a performance improvement over any of the individual views we construct and performance comparable with recent approaches in semi-supervised learning, but has some undesirable properties. To alleviate the issues present with co-training we present Meta Co-Training which is an extension of the successful Meta Pseudo Labels approach to two views. Our method achieves new state-of-the-art performance on ImageNet-10% with very few training resources, as well as outperforming prior semi-supervised work on several other fine-grained image classification datasets.