A common practice in transfer learning is to initialize the downstream model weights by pre-training on a data-abundant upstream task. In object detection specifically, the feature backbone is typically initialized with Imagenet classifier weights and fine-tuned on the object detection task. Recent works show this is not strictly necessary under longer training regimes and provide recipes for training the backbone from scratch. We investigate the opposite direction of this end-to-end training trend: we show that an extreme form of knowledge preservation -- freezing the classifier-initialized backbone -- consistently improves many different detection models, and leads to considerable resource savings. We hypothesize and corroborate experimentally that the remaining detector components capacity and structure is a crucial factor in leveraging the frozen backbone. Immediate applications of our findings include performance improvements on hard cases like detection of long-tail object classes and computational and memory resource savings that contribute to making the field more accessible to researchers with access to fewer computational resources.
Currently, this paper is under review in IEEE. Transformers have intrigued the vision research community with their state-of-the-art performance in natural language processing. With their superior performance, transformers have found their way in the field of hyperspectral image classification and achieved promising results. In this article, we harness the power of transformers to conquer the task of hyperspectral unmixing and propose a novel deep unmixing model with transformers. We aim to utilize the ability of transformers to better capture the global feature dependencies in order to enhance the quality of the endmember spectra and the abundance maps. The proposed model is a combination of a convolutional autoencoder and a transformer. The hyperspectral data is encoded by the convolutional encoder. The transformer captures long-range dependencies between the representations derived from the encoder. The data are reconstructed using a convolutional decoder. We applied the proposed unmixing model to three widely used unmixing datasets, i.e., Samson, Apex, and Washington DC mall and compared it with the state-of-the-art in terms of root mean squared error and spectral angle distance. The source code for the proposed model will be made publicly available at \url{https://github.com/preetam22n/DeepTrans-HSU}.
Vision transformers (ViTs) have been trending in image classification tasks due to their promising performance when compared to convolutional neural networks (CNNs). As a result, many researchers have tried to incorporate ViTs in hyperspectral image (HSI) classification tasks. To achieve satisfactory performance, close to that of CNNs, transformers need fewer parameters. ViTs and other similar transformers use an external classification (CLS) token which is randomly initialized and often fails to generalize well, whereas other sources of multimodal datasets, such as light detection and ranging (LiDAR) offer the potential to improve these models by means of a CLS. In this paper, we introduce a new multimodal fusion transformer (MFT) network which comprises a multihead cross patch attention (mCrossPA) for HSI land-cover classification. Our mCrossPA utilizes other sources of complementary information in addition to the HSI in the transformer encoder to achieve better generalization. The concept of tokenization is used to generate CLS and HSI patch tokens, helping to learn a {distinctive representation} in a reduced and hierarchical feature space. Extensive experiments are carried out on {widely used benchmark} datasets {i.e.,} the University of Houston, Trento, University of Southern Mississippi Gulfpark (MUUFL), and Augsburg. We compare the results of the proposed MFT model with other state-of-the-art transformers, classical CNNs, and conventional classifiers models. The superior performance achieved by the proposed model is due to the use of multihead cross patch attention. The source code will be made available publicly at \url{https://github.com/AnkurDeria/MFT}.}
Automatic image-based pavement distress detection and recognition are vital for pavement maintenance and management. However, existing deep learning-based methods largely omit the specific characteristics of pavement images, such as high image resolution and low distress area ratio, and are not end-to-end trainable. In this paper, we present a series of simple yet effective end-to-end deep learning approaches named Weakly Supervised Patch Label Inference Networks (WSPLIN) for efficiently addressing these tasks under various application settings. WSPLIN transforms the fully supervised pavement image classification problem into a weakly supervised pavement patch classification problem for solutions. Specifically, WSPLIN first divides the pavement image under different scales into patches with different collection strategies and then employs a Patch Label Inference Network (PLIN) to infer the labels of these patches to fully exploit the resolution and scale information. Notably, we design a patch label sparsity constraint based on the prior knowledge of distress distribution and leverage the Comprehensive Decision Network (CDN) to guide the training of PLIN in a weakly supervised way. Therefore, the patch labels produced by PLIN provide interpretable intermediate information, such as the rough location and the type of distress. We evaluate our method on a large-scale bituminous pavement distress dataset named CQU-BPDD and the augmented Crack500 (Crack500-PDD) dataset, which is a newly constructed pavement distress detection dataset augmented from the Crack500. Extensive results demonstrate the superiority of our method over baselines in both performance and efficiency. The source codes of WSPLIN are released on https://github.com/DearCaat/wsplin.
We introduce AmsterTime: a challenging dataset to benchmark visual place recognition (VPR) in presence of a severe domain shift. AmsterTime offers a collection of 2,500 well-curated images matching the same scene from a street view matched to historical archival image data from Amsterdam city. The image pairs capture the same place with different cameras, viewpoints, and appearances. Unlike existing benchmark datasets, AmsterTime is directly crowdsourced in a GIS navigation platform (Mapillary). We evaluate various baselines, including non-learning, supervised and self-supervised methods, pre-trained on different relevant datasets, for both verification and retrieval tasks. Our result credits the best accuracy to the ResNet-101 model pre-trained on the Landmarks dataset for both verification and retrieval tasks by 84% and 24%, respectively. Additionally, a subset of Amsterdam landmarks is collected for feature evaluation in a classification task. Classification labels are further used to extract the visual explanations using Grad-CAM for inspection of the learned similar visuals in a deep metric learning models.
Learning visual representation of high quality is essential for image classification. Recently, a series of contrastive representation learning methods have achieved preeminent success. Particularly, SupCon outperformed the dominant methods based on cross-entropy loss in representation learning. However, we notice that there could be potential ethical risks in supervised contrastive learning. In this paper, we for the first time analyze unfairness caused by supervised contrastive learning and propose a new Fair Supervised Contrastive Loss (FSCL) for fair visual representation learning. Inheriting the philosophy of supervised contrastive learning, it encourages representation of the same class to be closer to each other than that of different classes, while ensuring fairness by penalizing the inclusion of sensitive attribute information in representation. In addition, we introduce a group-wise normalization to diminish the disparities of intra-group compactness and inter-class separability between demographic groups that arouse unfair classification. Through extensive experiments on CelebA and UTK Face, we validate that the proposed method significantly outperforms SupCon and existing state-of-the-art methods in terms of the trade-off between top-1 accuracy and fairness. Moreover, our method is robust to the intensity of data bias and effectively works in incomplete supervised settings. Our code is available at https://github.com/sungho-CoolG/FSCL.
Grounded situation recognition is the task of predicting the main activity, entities playing certain roles within the activity, and bounding-box groundings of the entities in the given image. To effectively deal with this challenging task, we introduce a novel approach where the two processes for activity classification and entity estimation are interactive and complementary. To implement this idea, we propose Collaborative Glance-Gaze TransFormer (CoFormer) that consists of two modules: Glance transformer for activity classification and Gaze transformer for entity estimation. Glance transformer predicts the main activity with the help of Gaze transformer that analyzes entities and their relations, while Gaze transformer estimates the grounded entities by focusing only on the entities relevant to the activity predicted by Glance transformer. Our CoFormer achieves the state of the art in all evaluation metrics on the SWiG dataset. Training code and model weights are available at https://github.com/jhcho99/CoFormer.
In computer vision, it has achieved great transfer learning performance via adapting large-scale pretrained vision models (e.g., vision transformers) to downstream tasks. Common approaches for model adaptation either update all model parameters or leverage linear probes. In this paper, we aim to study parameter-efficient model adaptation strategies for vision transformers on the image classification task. We formulate efficient model adaptation as a subspace training problem and perform a comprehensive benchmarking over different efficient adaptation methods. We conduct an empirical study on each efficient model adaptation method focusing on its performance alongside parameter cost. Furthermore, we propose a parameter-efficient model adaptation framework, which first selects submodules by measuring local intrinsic dimensions and then projects them into subspace for further decomposition via a novel Kronecker Adaptation (KAdaptation) method. We analyze and compare our method with a diverse set of baseline model adaptation methods (including state-of-the-art methods for pretrained language models). Our method performs the best in terms of the tradeoff between accuracy and parameter efficiency across 20 image classification datasets under the few-shot setting and 7 image classification datasets under the full-shot setting.
The networks trained on the long-tailed dataset vary remarkably, despite the same training settings, which shows the great uncertainty in long-tailed learning. To alleviate the uncertainty, we propose a Nested Collaborative Learning (NCL), which tackles the problem by collaboratively learning multiple experts together. NCL consists of two core components, namely Nested Individual Learning (NIL) and Nested Balanced Online Distillation (NBOD), which focus on the individual supervised learning for each single expert and the knowledge transferring among multiple experts, respectively. To learn representations more thoroughly, both NIL and NBOD are formulated in a nested way, in which the learning is conducted on not just all categories from a full perspective but some hard categories from a partial perspective. Regarding the learning in the partial perspective, we specifically select the negative categories with high predicted scores as the hard categories by using a proposed Hard Category Mining (HCM). In the NCL, the learning from two perspectives is nested, highly related and complementary, and helps the network to capture not only global and robust features but also meticulous distinguishing ability. Moreover, self-supervision is further utilized for feature enhancement. Extensive experiments manifest the superiority of our method with outperforming the state-of-the-art whether by using a single model or an ensemble.
Currently, many theoretical as well as practically relevant questions towards the transferability and robustness of Convolutional Neural Networks (CNNs) remain unsolved. While ongoing research efforts are engaging these problems from various angles, in most computer vision related cases these approaches can be generalized to investigations of the effects of distribution shifts in image data. In this context, we propose to study the shifts in the learned weights of trained CNN models. Here we focus on the properties of the distributions of dominantly used 3x3 convolution filter kernels. We collected and publicly provide a dataset with over 1.4 billion filters from hundreds of trained CNNs, using a wide range of datasets, architectures, and vision tasks. In a first use case of the proposed dataset, we can show highly relevant properties of many publicly available pre-trained models for practical applications: I) We analyze distribution shifts (or the lack thereof) between trained filters along different axes of meta-parameters, like visual category of the dataset, task, architecture, or layer depth. Based on these results, we conclude that model pre-training can succeed on arbitrary datasets if they meet size and variance conditions. II) We show that many pre-trained models contain degenerated filters which make them less robust and less suitable for fine-tuning on target applications. Data & Project website: https://github.com/paulgavrikov/cnn-filter-db
In this paper, we consider image quality assessment (IQA) as a measure of how images are amenable with respect to a given downstream task, or task amenability. When the task is performed using machine learning algorithms, such as a neural-network-based task predictor for image classification or segmentation, the performance of the task predictor provides an objective estimate of task amenability. In this work, we use an IQA controller to predict the task amenability which, itself being parameterised by neural networks, can be trained simultaneously with the task predictor. We further develop a meta-reinforcement learning framework to improve the adaptability for both IQA controllers and task predictors, such that they can be fine-tuned efficiently on new datasets or meta-tasks. We demonstrate the efficacy of the proposed task-specific, adaptable IQA approach, using two clinical applications for ultrasound-guided prostate intervention and pneumonia detection on X-ray images.
Recently, the semantics of scene text has been proven to be essential in fine-grained image classification. However, the existing methods mainly exploit the literal meaning of scene text for fine-grained recognition, which might be irrelevant when it is not significantly related to objects/scenes. We propose an end-to-end trainable network that mines implicit contextual knowledge behind scene text image and enhance the semantics and correlation to fine-tune the image representation. Unlike the existing methods, our model integrates three modalities: visual feature extraction, text semantics extraction, and correlating background knowledge to fine-grained image classification. Specifically, we employ KnowBert to retrieve relevant knowledge for semantic representation and combine it with image features for fine-grained classification. Experiments on two benchmark datasets, Con-Text, and Drink Bottle, show that our method outperforms the state-of-the-art by 3.72\% mAP and 5.39\% mAP, respectively. To further validate the effectiveness of the proposed method, we create a new dataset on crowd activity recognition for the evaluation. The source code and new dataset of this work are available at https://github.com/lanfeng4659/KnowledgeMiningWithSceneText.
The question of whether pre-training on geometric tasks is viable for downstream transfer to semantic tasks is important for two reasons, one practical and the other scientific. If the answer is positive, we may be able to reduce pre-training cost and bias from human annotators significantly. If the answer is negative, it may shed light on the role of embodiment in the emergence of language and other cognitive functions in evolutionary history. To frame the question in a way that is testable with current means, we pre-train a model on a geometric task, and test whether that can be used to prime a notion of 'object' that enables inference of semantics as soon as symbols (labels) are assigned. We choose monocular depth prediction as the geometric task, and semantic segmentation as the downstream semantic task, and design a collection of empirical tests by exploring different forms of supervision, training pipelines, and data sources for both depth pre-training and semantic fine-tuning. We find that monocular depth is a viable form of pre-training for semantic segmentation, validated by improvements over common baselines. Based on the findings, we propose several possible mechanisms behind the improvements, including their relation to dataset size, resolution, architecture, in/out-of-domain source data, and validate them through a wide range of ablation studies. We also find that optical flow, which at first glance may seem as good as depth prediction since it optimizes the same photometric reprojection error, is considerably less effective, as it does not explicitly aim to infer the latent structure of the scene, but rather the raw phenomenology of temporally adjacent images.
Recently, vision-language pre-training shows great potential in open-vocabulary object detection, where detectors trained on base classes are devised for detecting new classes. The class text embedding is firstly generated by feeding prompts to the text encoder of a pre-trained vision-language model. It is then used as the region classifier to supervise the training of a detector. The key element that leads to the success of this model is the proper prompt, which requires careful words tuning and ingenious design. To avoid laborious prompt engineering, there are some prompt representation learning methods being proposed for the image classification task, which however can only be sub-optimal solutions when applied to the detection task. In this paper, we introduce a novel method, detection prompt (DetPro), to learn continuous prompt representations for open-vocabulary object detection based on the pre-trained vision-language model. Different from the previous classification-oriented methods, DetPro has two highlights: 1) a background interpretation scheme to include the proposals in image background into the prompt training; 2) a context grading scheme to separate proposals in image foreground for tailored prompt training. We assemble DetPro with ViLD, a recent state-of-the-art open-world object detector, and conduct experiments on the LVIS as well as transfer learning on the Pascal VOC, COCO, Objects365 datasets. Experimental results show that our DetPro outperforms the baseline ViLD in all settings, e.g., +3.4 APbox and +3.0 APmask improvements on the novel classes of LVIS. Code and models are available at https://github.com/dyabel/detpro.
The lack of adversarial robustness has been recognized as an important issue for state-of-the-art machine learning (ML) models, e.g., deep neural networks (DNNs). Thereby, robustifying ML models against adversarial attacks is now a major focus of research. However, nearly all existing defense methods, particularly for robust training, made the white-box assumption that the defender has the access to the details of an ML model (or its surrogate alternatives if available), e.g., its architectures and parameters. Beyond existing works, in this paper we aim to address the problem of black-box defense: How to robustify a black-box model using just input queries and output feedback? Such a problem arises in practical scenarios, where the owner of the predictive model is reluctant to share model information in order to preserve privacy. To this end, we propose a general notion of defensive operation that can be applied to black-box models, and design it through the lens of denoised smoothing (DS), a first-order (FO) certified defense technique. To allow the design of merely using model queries, we further integrate DS with the zeroth-order (gradient-free) optimization. However, a direct implementation of zeroth-order (ZO) optimization suffers a high variance of gradient estimates, and thus leads to ineffective defense. To tackle this problem, we next propose to prepend an autoencoder (AE) to a given (black-box) model so that DS can be trained using variance-reduced ZO optimization. We term the eventual defense as ZO-AE-DS. In practice, we empirically show that ZO-AE- DS can achieve improved accuracy, certified robustness, and query complexity over existing baselines. And the effectiveness of our approach is justified under both image classification and image reconstruction tasks. Codes are available at https://github.com/damon-demon/Black-Box-Defense.
Deep Neural Networks ( DNN s) are known to make overconfident mistakes, which makes their use problematic in safety-critical applications. State-of-the-art ( SOTA ) calibration techniques improve on the confidence of predicted labels alone and leave the confidence of non-max classes (e.g. top-2, top-5) uncalibrated. Such calibration is not suitable for label refinement using post-processing. Further, most SOTA techniques learn a few hyper-parameters post-hoc, leaving out the scope for image, or pixel specific calibration. This makes them unsuitable for calibration under domain shift, or for dense prediction tasks like semantic segmentation. In this paper, we argue for intervening at the train time itself, so as to directly produce calibrated DNN models. We propose a novel auxiliary loss function: Multi-class Difference in Confidence and Accuracy ( MDCA ), to achieve the same MDCA can be used in conjunction with other application/task-specific loss functions. We show that training with MDCA leads to better-calibrated models in terms of Expected Calibration Error ( ECE ), and Static Calibration Error ( SCE ) on image classification, and segmentation tasks. We report ECE ( SCE ) score of 0.72 (1.60) on the CIFAR 100 dataset, in comparison to 1.90 (1.71) by the SOTA. Under domain shift, a ResNet-18 model trained on PACS dataset using MDCA gives an average ECE ( SCE ) score of 19.7 (9.7) across all domains, compared to 24.2 (11.8) by the SOTA. For the segmentation task, we report a 2X reduction in calibration error on PASCAL - VOC dataset in comparison to Focal Loss. Finally, MDCA training improves calibration even on imbalanced data, and for natural language classification tasks. We have released the code here: code is available at https://github.com/mdca-loss
The pursuit algorithms integrated in multi-layer convolutional sparse coding (ML-CSC) can interpret the convolutional neural networks (CNNs). However, many current state-of-art (SOTA) pursuit algorithms require multiple iterations to optimize the solution of ML-CSC, which limits their applications to deeper CNNs due to high computational cost and large number of resources for getting very tiny gain of performance. In this study, we focus on the 0th iteration in pursuit algorithm by introducing an effective initialization strategy for each layer, by which the solution for ML-CSC can be improved. Specifically, we first propose a novel width-depth synchronous extension-based basis pursuit (WSEBP) algorithm which solves the ML-CSC problem without the limitation of the number of iterations compared to the SOTA algorithms and maximizes the performance by an effective initialization in each layer. Then, we propose a simple and unified ML-CSC-based classification network (ML-CSC-Net) which consists of an ML-CSC-based feature encoder and a fully-connected layer to validate the performance of WSEBP on image classification task. The experimental results show that our proposed WSEBP outperforms SOTA algorithms in terms of accuracy and consumption resources. In addition, the WSEBP integrated in CNNs can improve the performance of deeper CNNs and make them interpretable. Finally, taking VGG as an example, we propose WSEBP-VGG13 to enhance the performance of VGG13, which achieves competitive results on four public datasets, i.e., 87.79% vs. 86.83% on Cifar-10 dataset, 58.01% vs. 54.60% on Cifar-100 dataset, 91.52% vs. 89.58% on COVID-19 dataset, and 99.88% vs. 99.78% on Crack dataset, respectively. The results show the effectiveness of the proposed WSEBP, the improved performance of ML-CSC with WSEBP, and interpretation of the CNNs or deeper CNNs.
A fundamental and challenging problem in deep learning is catastrophic forgetting, i.e. the tendency of neural networks to fail to preserve the knowledge acquired from old tasks when learning new tasks. This problem has been widely investigated in the research community and several Incremental Learning (IL) approaches have been proposed in the past years. While earlier works in computer vision have mostly focused on image classification and object detection, more recently some IL approaches for semantic segmentation have been introduced. These previous works showed that, despite its simplicity, knowledge distillation can be effectively employed to alleviate catastrophic forgetting. In this paper, we follow this research direction and, inspired by recent literature on contrastive learning, we propose a novel distillation framework, Uncertainty-aware Contrastive Distillation (\method). In a nutshell, \method~is operated by introducing a novel distillation loss that takes into account all the images in a mini-batch, enforcing similarity between features associated to all the pixels from the same classes, and pulling apart those corresponding to pixels from different classes. In order to mitigate catastrophic forgetting, we contrast features of the new model with features extracted by a frozen model learned at the previous incremental step. Our experimental results demonstrate the advantage of the proposed distillation technique, which can be used in synergy with previous IL approaches, and leads to state-of-art performance on three commonly adopted benchmarks for incremental semantic segmentation. The code is available at \url{https://github.com/ygjwd12345/UCD}.
While class activation map (CAM) generated by image classification network has been widely used for weakly supervised object localization (WSOL) and semantic segmentation (WSSS), such classifiers usually focus on discriminative object regions. In this paper, we propose Contrastive learning for Class-agnostic Activation Map (C$^2$AM) generation only using unlabeled image data, without the involvement of image-level supervision. The core idea comes from the observation that i) semantic information of foreground objects usually differs from their backgrounds; ii) foreground objects with similar appearance or background with similar color/texture have similar representations in the feature space. We form the positive and negative pairs based on the above relations and force the network to disentangle foreground and background with a class-agnostic activation map using a novel contrastive loss. As the network is guided to discriminate cross-image foreground-background, the class-agnostic activation maps learned by our approach generate more complete object regions. We successfully extracted from C$^2$AM class-agnostic object bounding boxes for object localization and background cues to refine CAM generated by classification network for semantic segmentation. Extensive experiments on CUB-200-2011, ImageNet-1K, and PASCAL VOC2012 datasets show that both WSOL and WSSS can benefit from the proposed C$^2$AM.
We make contributions towards improving adaptive-optimizer performance. Our improvements are based on suppression of the range of adaptive stepsizes in the AdaBelief optimizer. Firstly, we show that the particular placement of the parameter epsilon within the update expressions of AdaBelief reduces the range of the adaptive stepsizes, making AdaBelief closer to SGD with momentum. Secondly, we extend AdaBelief by further suppressing the range of the adaptive stepsizes. To achieve the above goal, we perform mutual layerwise vector projections between the gradient g_t and its first momentum m_t before using them to estimate the second momentum. The new optimization method is referred to as Aida. Thirdly, extensive experimental results show that Aida outperforms nine optimizers when training transformers and LSTMs for NLP, and VGG and ResNet for image classification over CIAF10 and CIFAR100 while matching the best performance of the nine methods when training WGAN-GP models for image generation tasks. Furthermore, Aida produces higher validation accuracies than AdaBelief for training ResNet18 over ImageNet. Code is available <a href="https://github.com/guoqiang-x-zhang/AidaOptimizer">at this URL</a>