Panoptic segmentation involves a combination of joint semantic segmentation and instance segmentation, where image contents are divided into two types: things and stuff. We present Panoptic SegFormer, a general framework for panoptic segmentation with transformers. It contains three innovative components: an efficient deeply-supervised mask decoder, a query decoupling strategy, and an improved post-processing method. We also use Deformable DETR to efficiently process multi-scale features, which is a fast and efficient version of DETR. Specifically, we supervise the attention modules in the mask decoder in a layer-wise manner. This deep supervision strategy lets the attention modules quickly focus on meaningful semantic regions. It improves performance and reduces the number of required training epochs by half compared to Deformable DETR. Our query decoupling strategy decouples the responsibilities of the query set and avoids mutual interference between things and stuff. In addition, our post-processing strategy improves performance without additional costs by jointly considering classification and segmentation qualities to resolve conflicting mask overlaps. Our approach increases the accuracy 6.2\% PQ over the baseline DETR model. Panoptic SegFormer achieves state-of-the-art results on COCO test-dev with 56.2\% PQ. It also shows stronger zero-shot robustness over existing methods. The code is released at \url{https://github.com/zhiqi-li/Panoptic-SegFormer}.
Convolutional neural networks have enabled significant improvements in medical image-based diagnosis. It is, however, increasingly clear that these models are susceptible to performance degradation when facing spurious correlations and dataset shift, leading, e.g., to underperformance on underrepresented patient groups. In this paper, we compare two classification schemes on the ADNI MRI dataset: a simple logistic regression model using manually selected volumetric features, and a convolutional neural network trained on 3D MRI data. We assess the robustness of the trained models in the face of varying dataset splits, training set sex composition, and stage of disease. In contrast to earlier work in other imaging modalities, we do not observe a clear pattern of improved model performance for the majority group in the training dataset. Instead, while logistic regression is fully robust to dataset composition, we find that CNN performance is generally improved for both male and female subjects when including more female subjects in the training dataset. We hypothesize that this might be due to inherent differences in the pathology of the two sexes. Moreover, in our analysis, the logistic regression model outperforms the 3D CNN, emphasizing the utility of manual feature specification based on prior knowledge, and the need for more robust automatic feature selection.
The increasing popularity of attention mechanisms in deep learning algorithms for computer vision and natural language processing made these models attractive to other research domains. In healthcare, there is a strong need for tools that may improve the routines of the clinicians and the patients. Naturally, the use of attention-based algorithms for medical applications occurred smoothly. However, being healthcare a domain that depends on high-stake decisions, the scientific community must ponder if these high-performing algorithms fit the needs of medical applications. With this motto, this paper extensively reviews the use of attention mechanisms in machine learning (including Transformers) for several medical applications. This work distinguishes itself from its predecessors by proposing a critical analysis of the claims and potentialities of attention mechanisms presented in the literature through an experimental case study on medical image classification with three different use cases. These experiments focus on the integrating process of attention mechanisms into established deep learning architectures, the analysis of their predictive power, and a visual assessment of their saliency maps generated by post-hoc explanation methods. This paper concludes with a critical analysis of the claims and potentialities presented in the literature about attention mechanisms and proposes future research lines in medical applications that may benefit from these frameworks.
Cervical abnormal cell detection is a challenging task as the morphological discrepancies between abnormal and normal cells are usually subtle. To determine whether a cervical cell is normal or abnormal, cytopathologists always take surrounding cells as references to identify its abnormality. To mimic these behaviors, we propose to explore contextual relationships to boost the performance of cervical abnormal cell detection. Specifically, both contextual relationships between cells and cell-to-global images are exploited to enhance features of each region of interest (RoI) proposals. Accordingly, two modules, dubbed as RoI-relationship attention module (RRAM) and global RoI attention module (GRAM), are developed and their combination strategies are also investigated. We establish a strong baseline by using Double-Head Faster R-CNN with feature pyramid network (FPN) and integrate our RRAM and GRAM into it to validate the effectiveness of the proposed modules. Experiments conducted on a large cervical cell detection dataset reveal that the introduction of RRAM and GRAM both achieves better average precision (AP) than the baseline methods. Moreover, when cascading RRAM and GRAM, our method outperforms the state-of-the-art (SOTA) methods. Furthermore, we also show the proposed feature enhancing scheme can facilitate both image-level and smear-level classification. The code and trained models are publicly available at https://github.com/CVIU-CSU/CR4CACD.
Dynamic networks, e.g., Dynamic Convolution (DY-Conv) and the Mixture of Experts (MoE), have been extensively explored as they can considerably improve the model's representation power with acceptable computational cost. The common practice in implementing dynamic networks is to convert the given static layers into fully dynamic ones where all parameters are dynamic (at least within a single layer) and vary with the input. However, such a fully dynamic setting may cause redundant parameters and high deployment costs, limiting the applicability of dynamic networks to a broader range of tasks and models. The main contributions of our work are challenging the basic commonsense in dynamic networks and proposing a partially dynamic network, namely PAD-Net, to transform the redundant dynamic parameters into static ones. Also, we further design Iterative Mode Partition to partition dynamic and static parameters efficiently. Our method is comprehensively supported by large-scale experiments with two typical advanced dynamic architectures, i.e., DY-Conv and MoE, on both image classification and GLUE benchmarks. Encouragingly, we surpass the fully dynamic networks by $+0.7\%$ top-1 acc with only $30\%$ dynamic parameters for ResNet-50 and $+1.9\%$ average score in language understanding with only $50\%$ dynamic parameters for BERT. Code will be released at: \url{https://github.com/Shwai-He/PAD-Net}.
Domain shift widely exists in the visual world, while modern deep neural networks commonly suffer from severe performance degradation under domain shift due to the poor generalization ability, which limits the real-world applications. The domain shift mainly lies in the limited source environmental variations and the large distribution gap between source and unseen target data. To this end, we propose a unified framework, Style-HAllucinated Dual consistEncy learning (SHADE), to handle such domain shift in various visual tasks. Specifically, SHADE is constructed based on two consistency constraints, Style Consistency (SC) and Retrospection Consistency (RC). SC enriches the source situations and encourages the model to learn consistent representation across style-diversified samples. RC leverages general visual knowledge to prevent the model from overfitting to source data and thus largely keeps the representation consistent between the source and general visual models. Furthermore, we present a novel style hallucination module (SHM) to generate style-diversified samples that are essential to consistency learning. SHM selects basis styles from the source distribution, enabling the model to dynamically generate diverse and realistic samples during training. Extensive experiments demonstrate that our versatile SHADE can significantly enhance the generalization in various visual recognition tasks, including image classification, semantic segmentation and object detection, with different models, i.e., ConvNets and Transformer.
Reliable classification and detection of certain medical conditions, in images, with state-of-the-art semantic segmentation networks, require vast amounts of pixel-wise annotation. However, the public availability of such datasets is minimal. Therefore, semantic segmentation with image-level labels presents a promising alternative to this problem. Nevertheless, very few works have focused on evaluating this technique and its applicability to the medical sector. Due to their complexity and the small number of training examples in medical datasets, classifier-based weakly supervised networks like class activation maps (CAMs) struggle to extract useful information from them. However, most state-of-the-art approaches rely on them to achieve their improvements. Therefore, we propose a framework that can still utilize the low-quality CAM predictions of complicated datasets to improve the accuracy of our results. Our framework achieves that by first utilizing lower threshold CAMs to cover the target object with high certainty; second, by combining multiple low-threshold CAMs that even out their errors while highlighting the target object. We performed exhaustive experiments on the popular multi-modal BRATS and prostate DECATHLON segmentation challenge datasets. Using the proposed framework, we have demonstrated an improved dice score of up to 8% on BRATS and 6% on DECATHLON datasets compared to the previous state-of-the-art.
Spiking neural networks (SNNs) have shown advantages in computation and energy efficiency over traditional artificial neural networks (ANNs) thanks to their event-driven representations. SNNs also replace weight multiplications in ANNs with additions, which are more energy-efficient and less computationally intensive. However, it remains a challenge to train deep SNNs due to the discrete spike function. A popular approach to circumvent this challenge is ANN-to-SNN conversion. However, due to the quantization error and accumulating error, it often requires lots of time steps (high inference latency) to achieve high performance, which negates SNN's advantages. To this end, this paper proposes Fast-SNN that achieves high performance with low latency. We demonstrate the equivalent mapping between temporal quantization in SNNs and spatial quantization in ANNs, based on which the minimization of the quantization error is transferred to quantized ANN training. With the minimization of the quantization error, we show that the sequential error is the primary cause of the accumulating error, which is addressed by introducing a signed IF neuron model and a layer-wise fine-tuning mechanism. Our method achieves state-of-the-art performance and low latency on various computer vision tasks, including image classification, object detection, and semantic segmentation. Codes are available at: https://github.com/yangfan-hu/Fast-SNN.
Numerous pre-training techniques for visual document understanding (VDU) have recently shown substantial improvements in performance across a wide range of document tasks. However, these pre-trained VDU models cannot guarantee continued success when the distribution of test data differs from the distribution of training data. In this paper, to investigate how robust existing pre-trained VDU models are to various distribution shifts, we first develop an out-of-distribution (OOD) benchmark termed Do-GOOD for the fine-Grained analysis on Document image-related tasks specifically. The Do-GOOD benchmark defines the underlying mechanisms that result in different distribution shifts and contains 9 OOD datasets covering 3 VDU related tasks, e.g., document information extraction, classification and question answering. We then evaluate the robustness and perform a fine-grained analysis of 5 latest VDU pre-trained models and 2 typical OOD generalization algorithms on these OOD datasets. Results from the experiments demonstrate that there is a significant performance gap between the in-distribution (ID) and OOD settings for document images, and that fine-grained analysis of distribution shifts can reveal the brittle nature of existing pre-trained VDU models and OOD generalization algorithms. The code and datasets for our Do-GOOD benchmark can be found at https://github.com/MAEHCM/Do-GOOD.
Existing adversarial attacks against Object Detectors (ODs) suffer from two inherent limitations. Firstly, ODs have complicated meta-structure designs, hence most advanced attacks for ODs concentrate on attacking specific detector-intrinsic structures, which makes it hard for them to work on other detectors and motivates us to design a generic attack against ODs. Secondly, most works against ODs make Adversarial Examples (AEs) by generalizing image-level attacks from classification to detection, which brings redundant computations and perturbations in semantically meaningless areas (e.g., backgrounds) and leads to an emergency for seeking controllable attacks for ODs. To this end, we propose a generic white-box attack, LGP (local perturbations with adaptively global attacks), to blind mainstream object detectors with controllable perturbations. For a detector-agnostic attack, LGP tracks high-quality proposals and optimizes three heterogeneous losses simultaneously. In this way, we can fool the crucial components of ODs with a part of their outputs without the limitations of specific structures. Regarding controllability, we establish an object-wise constraint that exploits foreground-background separation adaptively to induce the attachment of perturbations to foregrounds. Experimentally, the proposed LGP successfully attacked sixteen state-of-the-art object detectors on MS-COCO and DOTA datasets, with promising imperceptibility and transferability obtained. Codes are publicly released in https://github.com/liguopeng0923/LGP.git
Computer vision using deep neural networks (DNNs) has brought about seminal changes in people's lives. Applications range from automotive, face recognition in the security industry, to industrial process monitoring. In some cases, DNNs infer even in safety-critical situations. Therefore, for practical applications, DNNs have to behave in a robust way to disturbances such as noise, pixelation, or blur. Blur directly impacts the performance of DNNs, which are often approximated as a disk-shaped kernel to model defocus. However, optics suggests that there are different kernel shapes depending on wavelength and location caused by optical aberrations. In practice, as the optical quality of a lens decreases, such aberrations increase. This paper proposes OpticsBench, a benchmark for investigating robustness to realistic, practically relevant optical blur effects. Each corruption represents an optical aberration (coma, astigmatism, spherical, trefoil) derived from Zernike Polynomials. Experiments on ImageNet show that for a variety of different pre-trained DNNs, the performance varies strongly compared to disk-shaped kernels, indicating the necessity of considering realistic image degradations. In addition, we show on ImageNet-100 with OpticsAugment that robustness can be increased by using optical kernels as data augmentation. Compared to a conventionally trained ResNeXt50, training with OpticsAugment achieves an average performance gain of 21.7% points on OpticsBench and 6.8% points on 2D common corruptions.
The analysis of Synthetic Aperture Radar (SAR) imagery is an important step in remote sensing applications, and it is a challenging problem due to its inherent speckle noise. One typical solution is to model the data using the $G_I^0$ distribution and extract its roughness information, which in turn can be used in posterior imaging tasks, such as segmentation, classification and interpretation. This leads to the need of quick and reliable estimation of the roughness parameter from SAR data, especially with high resolution images. Unfortunately, traditional parameter estimation procedures are slow and prone to estimation failures. In this work, we proposed a neural network-based estimation framework that first learns how to predict underlying parameters of $G_I^0$ samples and then can be used to estimate the roughness of unseen data. We show that this approach leads to an estimator that is quicker, yields less estimation error and is less prone to failures than the traditional estimation procedures for this problem, even when we use a simple network. More importantly, we show that this same methodology can be generalized to handle image inputs and, even if trained on purely synthetic data for a few seconds, is able to perform real time pixel-wise roughness estimation for high resolution real SAR imagery.
Owing to advancements in deep learning technology, Vision Transformers (ViTs) have demonstrated impressive performance in various computer vision tasks. Nonetheless, ViTs still face some challenges, such as high computational complexity and the absence of desirable inductive biases. To alleviate these issues, {the potential advantages of combining eagle vision with ViTs are explored. We summarize a Bi-Fovea Visual Interaction (BFVI) structure inspired by the unique physiological and visual characteristics of eagle eyes. A novel Bi-Fovea Self-Attention (BFSA) mechanism and Bi-Fovea Feedforward Network (BFFN) are proposed based on this structural design approach, which can be used to mimic the hierarchical and parallel information processing scheme of the biological visual cortex, enabling networks to learn feature representations of targets in a coarse-to-fine manner. Furthermore, a Bionic Eagle Vision (BEV) block is designed as the basic building unit based on the BFSA mechanism and BFFN. By stacking BEV blocks, a unified and efficient family of pyramid backbone networks called Eagle Vision Transformers (EViTs) is developed. Experimental results show that EViTs exhibit highly competitive performance in various computer vision tasks, such as image classification, object detection and semantic segmentation. Compared with other approaches, EViTs have significant advantages, especially in terms of performance and computational efficiency. Code is available at https://github.com/nkusyl/EViT
Road damage detection and assessment are crucial components of infrastructure maintenance. However, current methods often struggle with detecting multiple types of road damage in a single image, particularly at varying scales. This is due to the lack of road datasets with various damage types having varying scales. To overcome this deficiency, first, we present a novel dataset called Diverse Road Damage Dataset (DRDD) for road damage detection that captures the diverse road damage types in individual images, addressing a crucial gap in existing datasets. Then, we provide our model, RDD4D, that exploits Attention4D blocks, enabling better feature refinement across multiple scales. The Attention4D module processes feature maps through an attention mechanism combining positional encoding and "Talking Head" components to capture local and global contextual information. In our comprehensive experimental analysis comparing various state-of-the-art models on our proposed, our enhanced model demonstrated superior performance in detecting large-sized road cracks with an Average Precision (AP) of 0.458 and maintained competitive performance with an overall AP of 0.445. Moreover, we also provide results on the CrackTinyNet dataset; our model achieved around a 0.21 increase in performance. The code, model weights, dataset, and our results are available on \href{https://github.com/msaqib17/Road_Damage_Detection}{https://github.com/msaqib17/Road\_Damage\_Detection}.
Dynamic convolution enhances model capacity by adaptively combining multiple kernels, yet faces critical trade-offs: prior works either (1) incur significant parameter overhead by scaling kernel numbers linearly, (2) compromise inference speed through complex kernel interactions, or (3) struggle to jointly optimize dynamic attention and static kernels. We also observe that pre-trained Convolutional Neural Networks (CNNs) exhibit inter-layer redundancy akin to that in Large Language Models (LLMs). Specifically, dense convolutional layers can be efficiently replaced by derived ``child" layers generated from a shared ``parent" convolutional kernel through an adapter. To address these limitations and implement the weight-sharing mechanism, we propose a lightweight convolution kernel plug-in, named KernelDNA. It decouples kernel adaptation into input-dependent dynamic routing and pre-trained static modulation, ensuring both parameter efficiency and hardware-friendly inference. Unlike existing dynamic convolutions that expand parameters via multi-kernel ensembles, our method leverages cross-layer weight sharing and adapter-based modulation, enabling dynamic kernel specialization without altering the standard convolution structure. This design preserves the native computational efficiency of standard convolutions while enhancing representation power through input-adaptive kernel adjustments. Experiments on image classification and dense prediction tasks demonstrate that KernelDNA achieves state-of-the-art accuracy-efficiency balance among dynamic convolution variants. Our codes are available at https://github.com/haiduo/KernelDNA.
Over the past years, advances in artificial intelligence (AI) have demonstrated how AI can solve many perception and generation tasks, such as image classification and text writing, yet reasoning remains a challenge. This paper introduces the FLIP dataset, a benchmark for evaluating AI reasoning capabilities based on human verification tasks on the Idena blockchain. FLIP challenges present users with two orderings of 4 images, requiring them to identify the logically coherent one. By emphasizing sequential reasoning, visual storytelling, and common sense, FLIP provides a unique testbed for multimodal AI systems. Our experiments evaluate state-of-the-art models, leveraging both vision-language models (VLMs) and large language models (LLMs). Results reveal that even the best open-sourced and closed-sourced models achieve maximum accuracies of 75.5% and 77.9%, respectively, in zero-shot settings, compared to human performance of 95.3%. Captioning models aid reasoning models by providing text descriptions of images, yielding better results than when using the raw images directly, 69.6% vs. 75.2% for Gemini 1.5 Pro. Combining the predictions from 15 models in an ensemble increases the accuracy to 85.2%. These findings highlight the limitations of existing reasoning models and the need for robust multimodal benchmarks like FLIP. The full codebase and dataset will be available at https://github.com/aplesner/FLIP-Reasoning-Challenge.
We aim to develop a robust yet flexible visual foundation model for Earth observation. It should possess strong capabilities in recognizing and localizing diverse visual targets while providing compatibility with various input-output interfaces required across different task scenarios. Current systems cannot meet these requirements, as they typically utilize task-specific architecture trained on narrow data domains with limited semantic coverage. Our study addresses these limitations from two aspects: data and modeling. We first introduce an automatic data engine that enjoys significantly better scalability compared to previous human annotation or rule-based approaches. It has enabled us to create the largest dataset of its kind to date, comprising 270K image-text-mask triplets covering an unprecedented range of diverse semantic categories and attribute specifications. Based on this data foundation, we further propose a task unification paradigm that centers around referring expression segmentation. It effectively handles a wide range of vision-centric perception tasks, including classification, detection, segmentation, grounding, etc, using a single model without any task-specific heads. Combining these innovations on data and modeling, we present RemoteSAM, a foundation model that establishes new SoTA on several earth observation perception benchmarks, outperforming other foundation models such as Falcon, GeoChat, and LHRS-Bot with significantly higher efficiency. Models and data are publicly available at https://github.com/1e12Leon/RemoteSAM.
Novel computer vision architectures monopolize the spotlight, but the impact of the model architecture is often conflated with simultaneous changes to training methodology and scaling strategies. Our work revisits the canonical ResNet (He et al., 2015) and studies these three aspects in an effort to disentangle them. Perhaps surprisingly, we find that training and scaling strategies may matter more than architectural changes, and further, that the resulting ResNets match recent state-of-the-art models. We show that the best performing scaling strategy depends on the training regime and offer two new scaling strategies: (1) scale model depth in regimes where overfitting can occur (width scaling is preferable otherwise); (2) increase image resolution more slowly than previously recommended (Tan & Le, 2019). Using improved training and scaling strategies, we design a family of ResNet architectures, ResNet-RS, which are 1.7x - 2.7x faster than EfficientNets on TPUs, while achieving similar accuracies on ImageNet. In a large-scale semi-supervised learning setup, ResNet-RS achieves 86.2% top-1 ImageNet accuracy, while being 4.7x faster than EfficientNet NoisyStudent. The training techniques improve transfer performance on a suite of downstream tasks (rivaling state-of-the-art self-supervised algorithms) and extend to video classification on Kinetics-400. We recommend practitioners use these simple revised ResNets as baselines for future research.
We explore the value of weak labels in learning transferable representations for medical images. Compared to hand-labeled datasets, weak or inexact labels can be acquired in large quantities at significantly lower cost and can provide useful training signals for data-hungry models such as deep neural networks. We consider weak labels in the form of pseudo-labels and propose a semi-weakly supervised contrastive learning (SWCL) framework for representation learning using semi-weakly annotated images. Specifically, we train a semi-supervised model to propagate labels from a small dataset consisting of diverse image-level annotations to a large unlabeled dataset. Using the propagated labels, we generate a patch-level dataset for pretraining and formulate a multi-label contrastive learning objective to capture position-specific features encoded in each patch. We empirically validate the transfer learning performance of SWCL on seven public retinal fundus datasets, covering three disease classification tasks and two anatomical structure segmentation tasks. Our experiment results suggest that, under very low data regime, large-scale ImageNet pretraining on improved architecture remains a very strong baseline, and recently proposed self-supervised methods falter in segmentation tasks, possibly due to the strong invariant constraint imposed. Our method surpasses all prior self-supervised methods and standard cross-entropy training, while closing the gaps with ImageNet pretraining.
Transformer achieves remarkable successes in understanding 1 and 2-dimensional signals (e.g., NLP and Image Content Understanding). As a potential alternative to convolutional neural networks, it shares merits of strong interpretability, high discriminative power on hyper-scale data, and flexibility in processing varying length inputs. However, its encoders naturally contain computational intensive operations such as pair-wise self-attention, incurring heavy computational burden when being applied on the complex 3-dimensional video signals. This paper presents Token Shift Module (i.e., TokShift), a novel, zero-parameter, zero-FLOPs operator, for modeling temporal relations within each transformer encoder. Specifically, the TokShift barely temporally shifts partial [Class] token features back-and-forth across adjacent frames. Then, we densely plug the module into each encoder of a plain 2D vision transformer for learning 3D video representation. It is worth noticing that our TokShift transformer is a pure convolutional-free video transformer pilot with computational efficiency for video understanding. Experiments on standard benchmarks verify its robustness, effectiveness, and efficiency. Particularly, with input clips of 8/12 frames, the TokShift transformer achieves SOTA precision: 79.83%/80.40% on the Kinetics-400, 66.56% on EGTEA-Gaze+, and 96.80% on UCF-101 datasets, comparable or better than existing SOTA convolutional counterparts. Our code is open-sourced in: https://github.com/VideoNetworks/TokShift-Transformer.