Long-tailed datasets are very frequently encountered in real-world use cases where few classes or categories (known as majority or head classes) have higher number of data samples compared to the other classes (known as minority or tail classes). Training deep neural networks on such datasets gives results biased towards the head classes. So far, researchers have come up with multiple weighted loss and data re-sampling techniques in efforts to reduce the bias. However, most of such techniques assume that the tail classes are always the most difficult classes to learn and therefore need more weightage or attention. Here, we argue that the assumption might not always hold true. Therefore, we propose a novel approach to dynamically measure the instantaneous difficulty of each class during the training phase of the model. Further, we use the difficulty measures of each class to design a novel weighted loss technique called `class-wise difficulty based weighted (CDB-W) loss' and a novel data sampling technique called `class-wise difficulty based sampling (CDB-S)'. To verify the wide-scale usability of our CDB methods, we conducted extensive experiments on multiple tasks such as image classification, object detection, instance segmentation and video-action classification. Results verified that CDB-W loss and CDB-S could achieve state-of-the-art results on many class-imbalanced datasets such as ImageNet-LT, LVIS and EGTEA, that resemble real-world use cases.
In this work we introduce a novel medical image style transfer method, StyleMapper, that can transfer medical scans to an unseen style with access to limited training data. This is made possible by training our model on unlimited possibilities of simulated random medical imaging styles on the training set, making our work more computationally efficient when compared with other style transfer methods. Moreover, our method enables arbitrary style transfer: transferring images to styles unseen in training. This is useful for medical imaging, where images are acquired using different protocols and different scanner models, resulting in a variety of styles that data may need to be transferred between. Methods: Our model disentangles image content from style and can modify an image's style by simply replacing the style encoding with one extracted from a single image of the target style, with no additional optimization required. This also allows the model to distinguish between different styles of images, including among those that were unseen in training. We propose a formal description of the proposed model. Results: Experimental results on breast magnetic resonance images indicate the effectiveness of our method for style transfer. Conclusion: Our style transfer method allows for the alignment of medical images taken with different scanners into a single unified style dataset, allowing for the training of other downstream tasks on such a dataset for tasks such as classification, object detection and others.
Data-driven algorithms have surpassed traditional techniques in almost every aspect in robotic vision problems. Such algorithms need vast amounts of quality data to be able to work properly after their training process. Gathering and annotating that sheer amount of data in the real world is a time-consuming and error-prone task. Those problems limit scale and quality. Synthetic data generation has become increasingly popular since it is faster to generate and automatic to annotate. However, most of the current datasets and environments lack realism, interactions, and details from the real world. UnrealROX is an environment built over Unreal Engine 4 which aims to reduce that reality gap by leveraging hyperrealistic indoor scenes that are explored by robot agents which also interact with objects in a visually realistic manner in that simulated world. Photorealistic scenes and robots are rendered by Unreal Engine into a virtual reality headset which captures gaze so that a human operator can move the robot and use controllers for the robotic hands; scene information is dumped on a per-frame basis so that it can be reproduced offline to generate raw data and ground truth annotations. This virtual reality environment enables robotic vision researchers to generate realistic and visually plausible data with full ground truth for a wide variety of problems such as class and instance semantic segmentation, object detection, depth estimation, visual grasping, and navigation.
Recently, Vision Transformers (ViT), with the self-attention (SA) as the de facto ingredients, have demonstrated great potential in the computer vision community. For the sake of trade-off between efficiency and performance, a group of works merely perform SA operation within local patches, whereas the global contextual information is abandoned, which would be indispensable for visual recognition tasks. To solve the issue, the subsequent global-local ViTs take a stab at marrying local SA with global one in parallel or alternative way in the model. Nevertheless, the exhaustively combined local and global context may exist redundancy for various visual data, and the receptive field within each layer is fixed. Alternatively, a more graceful way is that global and local context can adaptively contribute per se to accommodate different visual data. To achieve this goal, we in this paper propose a novel ViT architecture, termed NomMer, which can dynamically Nominate the synergistic global-local context in vision transforMer. By investigating the working pattern of our proposed NomMer, we further explore what context information is focused. Beneficial from this "dynamic nomination" mechanism, without bells and whistles, the NomMer can not only achieve 84.5% Top-1 classification accuracy on ImageNet with only 73M parameters, but also show promising performance on dense prediction tasks, i.e., object detection and semantic segmentation. The code and models will be made publicly available at https://github.com/TencentYoutuResearch/VisualRecognition-NomMer
Although deep neural networks enable impressive visual perception performance for autonomous driving, their robustness to varying weather conditions still requires attention. When adapting these models for changed environments, such as different weather conditions, they are prone to forgetting previously learned information. This catastrophic forgetting is typically addressed via incremental learning approaches which usually re-train the model by either keeping a memory bank of training samples or keeping a copy of the entire model or model parameters for each scenario. While these approaches show impressive results, they can be prone to scalability issues and their applicability for autonomous driving in all weather conditions has not been shown. In this paper we propose DISC -- Domain Incremental through Statistical Correction -- a simple online zero-forgetting approach which can incrementally learn new tasks (i.e weather conditions) without requiring re-training or expensive memory banks. The only information we store for each task are the statistical parameters as we categorize each domain by the change in first and second order statistics. Thus, as each task arrives, we simply 'plug and play' the statistical vectors for the corresponding task into the model and it immediately starts to perform well on that task. We show the efficacy of our approach by testing it for object detection in a challenging domain-incremental autonomous driving scenario where we encounter different adverse weather conditions, such as heavy rain, fog, and snow.
Vision Transformers have shown great promise recently for many vision tasks due to the insightful architecture design and attention mechanism. By revisiting the self-attention responses in Transformers, we empirically observe two interesting issues. First, Vision Transformers present a queryirrelevant behavior at deep layers, where the attention maps exhibit nearly consistent contexts in global scope, regardless of the query patch position (also head-irrelevant). Second, the attention maps are intrinsically sparse, few tokens dominate the attention weights; introducing the knowledge from ConvNets would largely smooth the attention and enhance the performance. Motivated by above observations, we generalize self-attention formulation to abstract a queryirrelevant global context directly and further integrate the global context into convolutions. The resulting model, a Fully Convolutional Vision Transformer (i.e., FCViT), purely consists of convolutional layers and firmly inherits the merits of both attention mechanism and convolutions, including dynamic property, weight sharing, and short- and long-range feature modeling, etc. Experimental results demonstrate the effectiveness of FCViT. With less than 14M parameters, our FCViT-S12 outperforms related work ResT-Lite by 3.7% top1 accuracy on ImageNet-1K. When scaling FCViT to larger models, we still perform better than previous state-of-the-art ConvNeXt with even fewer parameters. FCViT-based models also demonstrate promising transferability to downstream tasks, like object detection, instance segmentation, and semantic segmentation. Codes and models are made available at: https://github.com/ma-xu/FCViT.
Accurate and efficient characterization of nanoparticles (NPs), particularly regarding particle size distribution, is essential for advancing our understanding of their structure-property relationships and facilitating their design for various applications. In this study, we introduce a novel two-stage artificial intelligence (AI)-driven workflow for NP analysis that leverages prompt engineering techniques from state-of-the-art single-stage object detection and large-scale vision transformer (ViT) architectures. This methodology was applied to transmission electron microscopy (TEM) and scanning TEM (STEM) images of heterogeneous catalysts, enabling high-resolution, high-throughput analysis of particle size distributions for supported metal catalysts. The model's performance in detecting and segmenting NPs was validated across diverse heterogeneous catalyst systems, including various metals (Cu, Ru, Pt, and PtCo), supports (silica ($\text{SiO}_2$), $\gamma$-alumina ($\gamma$-$\text{Al}_2\text{O}_3$), and carbon black), and particle diameter size distributions with means and standard deviations of 2.9 $\pm$ 1.1 nm, 1.6 $\pm$ 0.2 nm, 9.7 $\pm$ 4.6 nm, and 4 $\pm$ 1.0 nm. Additionally, the proposed machine learning (ML) approach successfully detects and segments overlapping NPs anchored on non-uniform catalytic support materials, providing critical insights into their spatial arrangements and interactions. Our AI-assisted NP analysis workflow demonstrates robust generalization across diverse datasets and can be readily applied to similar NP segmentation tasks without requiring costly model retraining.
The recent development of deep learning methods applied to vision has enabled their increasing integration into real-world applications to perform complex Computer Vision (CV) tasks. However, image acquisition conditions have a major impact on the performance of high-level image processing. A possible solution to overcome these limitations is to artificially augment the training databases or to design deep learning models that are robust to signal distortions. We opt here for the first solution by enriching the database with complex and realistic distortions which were ignored until now in the existing databases. To this end, we built a new versatile database derived from the well-known MS-COCO database to which we applied local and global photo-realistic distortions. These new local distortions are generated by considering the scene context of the images that guarantees a high level of photo-realism. Distortions are generated by exploiting the depth information of the objects in the scene as well as their semantics. This guarantees a high level of photo-realism and allows to explore real scenarios ignored in conventional databases dedicated to various CV applications. Our versatile database offers an efficient solution to improve the robustness of various CV tasks such as Object Detection (OD), scene segmentation, and distortion-type classification methods. The image database, scene classification index, and distortion generation codes are publicly available \footnote{\url{https://github.com/Aymanbegh/CD-COCO}}
This paper introduces an attacking mechanism to challenge the resilience of autonomous driving systems. Specifically, we manipulate the decision-making processes of an autonomous vehicle by dynamically displaying adversarial patches on a screen mounted on another moving vehicle. These patches are optimized to deceive the object detection models into misclassifying targeted objects, e.g., traffic signs. Such manipulation has significant implications for critical multi-vehicle interactions such as intersection crossing and lane changing, which are vital for safe and efficient autonomous driving systems. Particularly, we make four major contributions. First, we introduce a novel adversarial attack approach where the patch is not co-located with its target, enabling more versatile and stealthy attacks. Moreover, our method utilizes dynamic patches displayed on a screen, allowing for adaptive changes and movement, enhancing the flexibility and performance of the attack. To do so, we design a Screen Image Transformation Network (SIT-Net), which simulates environmental effects on the displayed images, narrowing the gap between simulated and real-world scenarios. Further, we integrate a positional loss term into the adversarial training process to increase the success rate of the dynamic attack. Finally, we shift the focus from merely attacking perceptual systems to influencing the decision-making algorithms of self-driving systems. Our experiments demonstrate the first successful implementation of such dynamic adversarial attacks in real-world autonomous driving scenarios, paving the way for advancements in the field of robust and secure autonomous driving.
Advances in multi-modal embeddings, and in particular CLIP, have recently driven several breakthroughs in Computer Vision (CV). CLIP has shown impressive performance on a variety of tasks, yet, its inherently opaque architecture may hinder the application of models employing CLIP as backbone, especially in fields where trust and model explainability are imperative, such as in the medical domain. Current explanation methodologies for CV models rely on Saliency Maps computed through gradient analysis or input perturbation. However, these Saliency Maps can only be computed to explain classes relevant to the end task, often smaller in scope than the backbone training classes. In the context of models implementing CLIP as their vision backbone, a substantial portion of the information embedded within the learned representations is thus left unexplained. In this work, we propose Concept Visualization (ConVis), a novel saliency methodology that explains the CLIP embedding of an image by exploiting the multi-modal nature of the embeddings. ConVis makes use of lexical information from WordNet to compute task-agnostic Saliency Maps for any concept, not limited to concepts the end model was trained on. We validate our use of WordNet via an out of distribution detection experiment, and test ConVis on an object localization benchmark, showing that Concept Visualizations correctly identify and localize the image's semantic content. Additionally, we perform a user study demonstrating that our methodology can give users insight on the model's functioning.
Multi-modality image fusion aims at fusing modality-specific (complementarity) and modality-shared (correlation) information from multiple source images. To tackle the problem of the neglect of inter-feature relationships, high-frequency information loss, and the limited attention to downstream tasks, this paper focuses on how to model correlation-driven decomposing features and reason high-level graph representation by efficiently extracting complementary information and aggregating multi-guided features. We propose a three-branch encoder-decoder architecture along with corresponding fusion layers as the fusion strategy. Firstly, shallow features from individual modalities are extracted by a depthwise convolution layer combined with the transformer block. In the three parallel branches of the encoder, Cross Attention and Invertible Block (CAI) extracts local features and preserves high-frequency texture details. Base Feature Extraction Module (BFE) captures long-range dependencies and enhances modality-shared information. Graph Reasoning Module (GR) is introduced to reason high-level cross-modality relations and simultaneously extract low-level detail features as CAI's modality-specific complementary information. Experiments demonstrate the competitive results compared with state-of-the-art methods in visible/infrared image fusion and medical image fusion tasks. Moreover, the proposed algorithm surpasses the state-of-the-art methods in terms of subsequent tasks, averagely scoring 8.27% [email omitted] higher in object detection and 5.85% mIoU higher in semantic segmentation. The code is avaliable at https://github.com/Abraham-Einstein/SMFNet/.
Neural Architecture Search (NAS) has shown great success in automating the design of neural networks, but the prohibitive amount of computations behind current NAS methods requires further investigations in improving the sample efficiency and the network evaluation cost to get better results in a shorter time. In this paper, we present a novel scalable Monte Carlo Tree Search (MCTS) based NAS agent, named AlphaX, to tackle these two aspects. AlphaX improves the search efficiency by adaptively balancing the exploration and exploitation at the state level, and by a Meta-Deep Neural Network (DNN) to predict network accuracies for biasing the search toward a promising region. To amortize the network evaluation cost, AlphaX accelerates MCTS rollouts with a distributed design and reduces the number of epochs in evaluating a network by transfer learning guided with the tree structure in MCTS. In 12 GPU days and 1000 samples, AlphaX found an architecture that reaches 97.84\% top-1 accuracy on CIFAR-10, and 75.5\% top-1 accuracy on ImageNet, exceeding SOTA NAS methods in both the accuracy and sampling efficiency. Particularly, we also evaluate AlphaX on NASBench-101, a large scale NAS dataset; AlphaX is 3x and 2.8x more sample efficient than Random Search and Regularized Evolution in finding the global optimum. Finally, we show the searched architecture improves a variety of vision applications from Neural Style Transfer, to Image Captioning and Object Detection.
Tracking players in sports videos is commonly done in a tracking-by-detection framework, first detecting players in each frame, and then performing association over time. While for some sports tracking players is sufficient for game analysis, sports like hockey, tennis and polo may require additional detections, that include the object the player is holding (e.g. racket, stick). The baseline solution for this problem involves detecting these objects as separate classes, and matching them to player detections based on the intersection over union (IoU). This approach, however, leads to poor matching performance in crowded situations, as it does not model the relationship between players and objects. In this paper, we propose a simple yet efficient way to detect and match players and related objects at once without extra cost, by considering an implicit association for prediction of multiple objects through the same proposal box. We evaluate the method on a dataset of broadcast ice hockey videos, and also a new public dataset we introduce called COCO +Torso. On the ice hockey dataset, the proposed method boosts matching performance from 57.1% to 81.4%, while also improving the meanAP of player+stick detections from 68.4% to 88.3%. On the COCO +Torso dataset, we see matching improving from 47.9% to 65.2%. The COCO +Torso dataset, code and pre-trained models will be released at https://github.com/foreverYoungGitHub/detect-and-match-related-objects.
Convolution-BatchNorm (ConvBN) blocks are integral components in various computer vision tasks and other domains. A ConvBN block can operate in three modes: Train, Eval, and Deploy. While the Train mode is indispensable for training models from scratch, the Eval mode is suitable for transfer learning and beyond, and the Deploy mode is designed for the deployment of models. This paper focuses on the trade-off between stability and efficiency in ConvBN blocks: Deploy mode is efficient but suffers from training instability; Eval mode is widely used in transfer learning but lacks efficiency. To solve the dilemma, we theoretically reveal the reason behind the diminished training stability observed in the Deploy mode. Subsequently, we propose a novel Tune mode to bridge the gap between Eval mode and Deploy mode. The proposed Tune mode is as stable as Eval mode for transfer learning, and its computational efficiency closely matches that of the Deploy mode. Through extensive experiments in object detection, classification, and adversarial example generation across $5$ datasets and $12$ model architectures, we demonstrate that the proposed Tune mode retains the performance while significantly reducing GPU memory footprint and training time, thereby contributing efficient ConvBN blocks for transfer learning and beyond. Our method has been integrated into both PyTorch (general machine learning framework) and MMCV/MMEngine (computer vision framework). Practitioners just need one line of code to enjoy our efficient ConvBN blocks thanks to PyTorch's builtin machine learning compilers.
Recently, efficient Vision Transformers have shown great performance with low latency on resource-constrained devices. Conventionally, they use 4x4 patch embeddings and a 4-stage structure at the macro level, while utilizing sophisticated attention with multi-head configuration at the micro level. This paper aims to address computational redundancy at all design levels in a memory-efficient manner. We discover that using larger-stride patchify stem not only reduces memory access costs but also achieves competitive performance by leveraging token representations with reduced spatial redundancy from the early stages. Furthermore, our preliminary analyses suggest that attention layers in the early stages can be substituted with convolutions, and several attention heads in the latter stages are computationally redundant. To handle this, we introduce a single-head attention module that inherently prevents head redundancy and simultaneously boosts accuracy by parallelly combining global and local information. Building upon our solutions, we introduce SHViT, a Single-Head Vision Transformer that obtains the state-of-the-art speed-accuracy tradeoff. For example, on ImageNet-1k, our SHViT-S4 is 3.3x, 8.1x, and 2.4x faster than MobileViTv2 x1.0 on GPU, CPU, and iPhone12 mobile device, respectively, while being 1.3% more accurate. For object detection and instance segmentation on MS COCO using Mask-RCNN head, our model achieves performance comparable to FastViT-SA12 while exhibiting 3.8x and 2.0x lower backbone latency on GPU and mobile device, respectively.
CLIP (Contrastive Language-Image Pre-training) uses contrastive learning from noise image-text pairs to excel at recognizing a wide array of candidates, yet its focus on broad associations hinders the precision in distinguishing subtle differences among fine-grained items. Conversely, Multimodal Large Language Models (MLLMs) excel at classifying fine-grained categories, thanks to their substantial knowledge from pre-training on web-level corpora. However, the performance of MLLMs declines with an increase in category numbers, primarily due to growing complexity and constraints of limited context window size. To synergize the strengths of both approaches and enhance the few-shot/zero-shot recognition abilities for datasets characterized by extensive and fine-grained vocabularies, this paper introduces RAR, a Retrieving And Ranking augmented method for MLLMs. We initially establish a multi-modal retriever based on CLIP to create and store explicit memory for different categories beyond the immediate context window. During inference, RAR retrieves the top-k similar results from the memory and uses MLLMs to rank and make the final predictions. Our proposed approach not only addresses the inherent limitations in fine-grained recognition but also preserves the model's comprehensive knowledge base, significantly boosting accuracy across a range of vision-language recognition tasks. Notably, our approach demonstrates a significant improvement in performance on 5 fine-grained visual recognition benchmarks, 11 few-shot image recognition datasets, and the 2 object detection datasets under the zero-shot recognition setting.
Cell tracking is a ubiquitous image analysis task in live-cell microscopy. Unlike multiple object tracking (MOT) for natural images, cell tracking typically involves hundreds of similar-looking objects that can divide in each frame, making it a particularly challenging problem. Current state-of-the-art approaches follow the tracking-by-detection paradigm, i.e. first all cells are detected per frame and successively linked in a second step to form biologically consistent cell tracks. Linking is commonly solved via discrete optimization methods, which require manual tuning of hyperparameters for each dataset and are therefore cumbersome to use in practice. Here we propose Trackastra, a general purpose cell tracking approach that uses a simple transformer architecture to directly learn pairwise associations of cells within a temporal window from annotated data. Importantly, unlike existing transformer-based MOT pipelines, our learning architecture also accounts for dividing objects such as cells and allows for accurate tracking even with simple greedy linking, thus making strides towards removing the requirement for a complex linking step. The proposed architecture operates on the full spatio-temporal context of detections within a time window by avoiding the computational burden of processing dense images. We show that our tracking approach performs on par with or better than highly tuned state-of-the-art cell tracking algorithms for various biological datasets, such as bacteria, cell cultures and fluorescent particles. We provide code at https://github.com/weigertlab/trackastra.
Contrastive language image pretraining (CLIP) encoders have been shown to be beneficial for a range of visual tasks from classification and detection to captioning and image manipulation. We investigate the effectiveness of CLIP visual backbones for Embodied AI tasks. We build incredibly simple baselines, named EmbCLIP, with no task specific architectures, inductive biases (such as the use of semantic maps), auxiliary tasks during training, or depth maps -- yet we find that our improved baselines perform very well across a range of tasks and simulators. EmbCLIP tops the RoboTHOR ObjectNav leaderboard by a huge margin of 20 pts (Success Rate). It tops the iTHOR 1-Phase Rearrangement leaderboard, beating the next best submission, which employs Active Neural Mapping, and more than doubling the % Fixed Strict metric (0.08 to 0.17). It also beats the winners of the 2021 Habitat ObjectNav Challenge, which employ auxiliary tasks, depth maps, and human demonstrations, and those of the 2019 Habitat PointNav Challenge. We evaluate the ability of CLIP's visual representations at capturing semantic information about input observations -- primitives that are useful for navigation-heavy embodied tasks -- and find that CLIP's representations encode these primitives more effectively than ImageNet-pretrained backbones. Finally, we extend one of our baselines, producing an agent capable of zero-shot object navigation that can navigate to objects that were not used as targets during training. Our code and models are available at https://github.com/allenai/embodied-clip
Do we on the right way for remote sensing image understanding (RSIU) by training models via supervised data-dependent and task-dependent way, instead of human vision in a label-free and task-independent way? We argue that a more desirable RSIU model should be trained with intrinsic structure from data rather that extrinsic human labels to realize generalizability across a wide range of RSIU tasks. According to this hypothesis, we proposed \textbf{T}he \textbf{O}riginal \textbf{V}ision model (TOV) in remote sensing filed. Trained by massive unlabeled optical data along a human-like self-supervised learning (SSL) path that is from general knowledge to specialized knowledge, TOV model can be easily adapted to various RSIU tasks, including scene classification, object detection, and semantic segmentation, and outperforms dominant ImageNet supervised pretrained method as well as two recently proposed SSL pretrained methods on majority of 12 publicly available benchmarks. Moreover, we analyze the influences of two key factors on the performance of building TOV model for RSIU, including the influence of using different data sampling methods and the selection of learning paths during self-supervised optimization. We believe that a general model which is trained by a label-free and task-independent way may be the next paradigm for RSIU and hope the insights distilled from this study can help to foster the development of an original vision model for RSIU.
Attention within windows has been widely explored in vision transformers to balance the performance, computation complexity, and memory footprint. However, current models adopt a hand-crafted fixed-size window design, which restricts their capacity of modeling long-term dependencies and adapting to objects of different sizes. To address this drawback, we propose \textbf{V}aried-\textbf{S}ize Window \textbf{A}ttention (VSA) to learn adaptive window configurations from data. Specifically, based on the tokens within each default window, VSA employs a window regression module to predict the size and location of the target window, i.e., the attention area where the key and value tokens are sampled. By adopting VSA independently for each attention head, it can model long-term dependencies, capture rich context from diverse windows, and promote information exchange among overlapped windows. VSA is an easy-to-implement module that can replace the window attention in state-of-the-art representative models with minor modifications and negligible extra computational cost while improving their performance by a large margin, e.g., 1.1\% for Swin-T on ImageNet classification. In addition, the performance gain increases when using larger images for training and test. Experimental results on more downstream tasks, including object detection, instance segmentation, and semantic segmentation, further demonstrate the superiority of VSA over the vanilla window attention in dealing with objects of different sizes. The code will be released https://github.com/ViTAE-Transformer/ViTAE-VSA.