Most semi-supervised learning methods over-sample labeled data when constructing training mini-batches. This paper studies whether this common practice improves learning and how. We compare it to an alternative setting where each mini-batch is uniformly sampled from all the training data, labeled or not, which greatly reduces direct supervision from true labels in typical low-label regimes. However, this simpler setting can also be seen as more general and even necessary in multi-task problems where over-sampling labeled data would become intractable. Our experiments on semi-supervised CIFAR-10 image classification using FixMatch show a performance drop when using the uniform sampling approach which diminishes when the amount of labeled data or the training time increases. Further, we analyse the training dynamics to understand how over-sampling of labeled data compares to uniform sampling. Our main finding is that over-sampling is especially beneficial early in training but gets less important in the later stages when more pseudo-labels become correct. Nevertheless, we also find that keeping some true labels remains important to avoid the accumulation of confirmation errors from incorrect pseudo-labels.
Transformers have recently shown superior performances on various vision tasks. The large, sometimes even global, receptive field endows Transformer models with higher representation power over their CNN counterparts. Nevertheless, simply enlarging receptive field also gives rise to several concerns. On the one hand, using dense attention e.g., in ViT, leads to excessive memory and computational cost, and features can be influenced by irrelevant parts which are beyond the region of interests. On the other hand, the sparse attention adopted in PVT or Swin Transformer is data agnostic and may limit the ability to model long range relations. To mitigate these issues, we propose a novel deformable self-attention module, where the positions of key and value pairs in self-attention are selected in a data-dependent way. This flexible scheme enables the self-attention module to focus on relevant regions and capture more informative features. On this basis, we present Deformable Attention Transformer, a general backbone model with deformable attention for both image classification and dense prediction tasks. Extensive experiments show that our models achieve consistently improved results on comprehensive benchmarks. Code is available at https://github.com/LeapLabTHU/DAT.
As a representation learning method, nearest regularized subspace(NRS) algorithm is an effective tool to obtain both accuracy and speed for PolSAR image classification. However, existing NRS methods use the polarimetric feature vector but the PolSAR original covariance matrix(known as Hermitian positive definite(HPD)matrix) as the input. Without considering the matrix structure, existing NRS-based methods cannot learn correlation among channels. How to utilize the original covariance matrix to NRS method is a key problem. To address this limit, a Riemannian NRS method is proposed, which consider the HPD matrices endow in the Riemannian space. Firstly, to utilize the PolSAR original data, a Riemannian NRS method(RNRS) is proposed by constructing HPD dictionary and HPD distance metric. Secondly, a new Tikhonov regularization term is designed to reduce the differences within the same class. Finally, the optimal method is developed and the first-order derivation is inferred. During the experimental test, only T matrix is used in the proposed method, while multiple of features are utilized for compared methods. Experimental results demonstrate the proposed method can outperform the state-of-art algorithms even using less features.
Large-scale image databases remain largely biased towards objects and activities encountered in a select few cultures. This absence of culturally-diverse images, which we refer to as the hidden tail, limits the applicability of pre-trained neural networks and inadvertently excludes researchers from under-represented regions. To begin remedying this issue, we curate Turath-150K, a database of images of the Arab world that reflect objects, activities, and scenarios commonly found there. In the process, we introduce three benchmark databases, Turath Standard, Art, and UNESCO, specialised subsets of the Turath dataset. After demonstrating the limitations of existing networks pre-trained on ImageNet when deployed on such benchmarks, we train and evaluate several networks on the task of image classification. As a consequence of Turath, we hope to engage machine learning researchers in under-represented regions, and to inspire the release of additional culture-focused databases. The database can be accessed here: danikiyasseh.github.io/Turath.
Machine learning systems often experience a distribution shift between training and testing. In this paper, we introduce a simple variational objective whose optima are exactly the set of all representations on which risk minimizers are guaranteed to be robust to any distribution shift that preserves the Bayes predictor, e.g., covariate shifts. Our objective has two components. First, a representation must remain discriminative for the task, i.e., some predictor must be able to simultaneously minimize the source and target risk. Second, the representation's marginal support needs to be the same across source and target. We make this practical by designing self-supervised objectives that only use unlabelled data and augmentations to train robust representations. Our objectives give insights into the robustness of CLIP, and further improve CLIP's representations to achieve SOTA results on DomainBed.
This paper quantifies the quality of heatmap-based eXplainable AI (XAI) methods w.r.t image classification problem. Here, a heatmap is considered desirable if it improves the probability of predicting the correct classes. Different XAI heatmap-based methods are empirically shown to improve classification confidence to different extents depending on the datasets, e.g. Saliency works best on ImageNet and Deconvolution on Chest X-Ray Pneumonia dataset. The novelty includes a new gap distribution that shows a stark difference between correct and wrong predictions. Finally, the generative augmentative explanation is introduced, a method to generate heatmaps capable of improving predictive confidence to a high level.
Quantized neural networks typically require smaller memory footprints and lower computation complexity, which is crucial for efficient deployment. However, quantization inevitably leads to a distribution divergence from the original network, which generally degrades the performance. To tackle this issue, massive efforts have been made, but most existing approaches lack statistical considerations and depend on several manual configurations. In this paper, we present an adaptive-mapping quantization method to learn an optimal latent sub-distribution that is inherent within models and smoothly approximated with a concrete Gaussian Mixture (GM). In particular, the network weights are projected in compliance with the GM-approximated sub-distribution. This sub-distribution evolves along with the weight update in a co-tuning schema guided by the direct task-objective optimization. Sufficient experiments on image classification and object detection over various modern architectures demonstrate the effectiveness, generalization property, and transferability of the proposed method. Besides, an efficient deployment flow for the mobile CPU is developed, achieving up to 7.46$\times$ inference acceleration on an octa-core ARM CPU. Our codes have been publicly released at \url{https://github.com/RunpeiDong/DGMS}.
Using natural language as a supervision for training visual recognition models holds great promise. Recent works have shown that if such supervision is used in the form of alignment between images and captions in large training datasets, then the resulting aligned models perform well on zero-shot classification as downstream tasks2. In this paper, we focus on teasing out what parts of the language supervision are essential for training zero-shot image classification models. Through extensive and careful experiments, we show that: 1) A simple Bag-of-Words (BoW) caption could be used as a replacement for most of the image captions in the dataset. Surprisingly, we observe that this approach improves the zero-shot classification performance when combined with word balancing. 2) Using a BoW pretrained model, we can obtain more training data by generating pseudo-BoW captions on images that do not have a caption. Models trained on images with real and pseudo-BoW captions achieve stronger zero-shot performance. On ImageNet-1k zero-shot evaluation, our best model, that uses only 3M image-caption pairs, performs on-par with a CLIP model trained on 15M image-caption pairs (31.5% vs 31.3%).
Recently, open-vocabulary image classification by vision language pre-training has demonstrated incredible achievements, that the model can classify arbitrary categories without seeing additional annotated images of that category. However, it is still unclear how to make the open-vocabulary recognition work well on broader vision problems. This paper targets open-vocabulary semantic segmentation by building it on an off-the-shelf pre-trained vision-language model, i.e., CLIP. However, semantic segmentation and the CLIP model perform on different visual granularity, that semantic segmentation processes on pixels while CLIP performs on images. To remedy the discrepancy in processing granularity, we refuse the use of the prevalent one-stage FCN based framework, and advocate a two-stage semantic segmentation framework, with the first stage extracting generalizable mask proposals and the second stage leveraging an image based CLIP model to perform open-vocabulary classification on the masked image crops which are generated in the first stage. Our experimental results show that this two-stage framework can achieve superior performance than FCN when trained only on COCO Stuff dataset and evaluated on other datasets without fine-tuning. Moreover, this simple framework also surpasses previous state-of-the-arts of zero-shot semantic segmentation by a large margin: +29.5 hIoU on the Pascal VOC 2012 dataset, and +8.9 hIoU on the COCO Stuff dataset. With its simplicity and strong performance, we hope this framework to serve as a baseline to facilitate future research. The code are made publicly available at~\url{https://github.com/MendelXu/zsseg.baseline}.
By leveraging deep learning to automatically classify camera trap images, ecologists can monitor biodiversity conservation efforts and the effects of climate change on ecosystems more efficiently. Due to the imbalanced class-distribution of camera trap datasets, current models are biased towards the majority classes. As a result, they obtain good performance for a few majority classes but poor performance for many minority classes. We used two-phase training to increase the performance for these minority classes. We trained, next to a baseline model, four models that implemented a different versions of two-phase training on a subset of the highly imbalanced Snapshot Serengeti dataset. Our results suggest that two-phase training can improve performance for many minority classes, with limited loss in performance for the other classes. We find that two-phase training based on majority undersampling increases class-specific F1-scores up to 3.0%. We also find that two-phase training outperforms using only oversampling or undersampling by 6.1% in F1-score on average. Finally, we find that a combination of over- and undersampling leads to a better performance than using them individually.
Autonomous systems are highly vulnerable to a variety of adversarial attacks on Deep Neural Networks (DNNs). Training-free model-agnostic defenses have recently gained popularity due to their speed, ease of deployment, and ability to work across many DNNs. To this end, a new technique has emerged for mitigating attacks on image classification DNNs, namely, preprocessing adversarial images using super resolution -- upscaling low-quality inputs into high-resolution images. This defense requires running both image classifiers and super resolution models on constrained autonomous systems. However, super resolution incurs a heavy computational cost. Therefore, in this paper, we investigate the following question: Does the robustness of image classifiers suffer if we use tiny super resolution models? To answer this, we first review a recent work called Super-Efficient Super Resolution (SESR) that achieves similar or better image quality than prior art while requiring 2x to 330x fewer Multiply-Accumulate (MAC) operations. We demonstrate that despite being orders of magnitude smaller than existing models, SESR achieves the same level of robustness as significantly larger networks. Finally, we estimate end-to-end performance of super resolution-based defenses on a commercial Arm Ethos-U55 micro-NPU. Our findings show that SESR achieves nearly 3x higher FPS than a baseline while achieving similar robustness.
Recently, Vision Transformer (ViT) has continuously established new milestones in the computer vision field, while the high computation and memory cost makes its propagation in industrial production difficult. Pruning, a traditional model compression paradigm for hardware efficiency, has been widely applied in various DNN structures. Nevertheless, it stays ambiguous on how to perform exclusive pruning on the ViT structure. Considering three key points: the structural characteristics, the internal data pattern of ViTs, and the related edge device deployment, we leverage the input token sparsity and propose a computation-aware soft pruning framework, which can be set up on vanilla Transformers of both flatten and CNN-type structures, such as Pooling-based ViT (PiT). More concretely, we design a dynamic attention-based multi-head token selector, which is a lightweight module for adaptive instance-wise token selection. We further introduce a soft pruning technique, which integrates the less informative tokens generated by the selector module into a package token that will participate in subsequent calculations rather than being completely discarded. Our framework is bound to the trade-off between accuracy and computation constraints of specific edge devices through our proposed computation-aware training strategy. Experimental results show that our framework significantly reduces the computation cost of ViTs while maintaining comparable performance on image classification. Moreover, our framework can guarantee the identified model to meet resource specifications of mobile devices and FPGA, and even achieve the real-time execution of DeiT-T on mobile platforms. For example, our method reduces the latency of DeiT-T to 26 ms (26%$\sim $41% superior to existing works) on the mobile device with 0.25%$\sim $4% higher top-1 accuracy on ImageNet.
Recently, the semi-supervised graph convolutional network (SSGCN) has been verified effective for hyperspectral image (HSI) classification. However, constrained by the limited training data and spectral uncertainty, the classification performance is remained to be further improved. Moreover, attribute to the massive data, the SSGCN with complex computation is generally too time- and resource-consuming to be applicable in real-time needs. To conquer these issues, we propose an efficient symmetric graph metric learning (SGML) framework by incorporating metric learning into the SSGCN paradigm. Specifically, we first conduct multilevel pixel-to-superpixel projection (P-SP) on the HSI to investigate the multiscale spatial information, where the suitable superpixel numbers are adaptively determined. Then, to extract more expressive representations, we design a new structure denoted as GSvolution, comprising the graph convolution (G-Conv) and a novel self-channel-enhanced convolution (S-Conv), to propagate the labeled and unlabeled graph node information and simultaneously enhance the critical intranode channel features. Finally, the superpixel node features are reprojected to the pixel level (SP-P) so that the distilled multistream features can be integrated to obtain the final decision. Noticeably, this ingenious symmetric mechanism (P-SP and SP-P) can alleviate the spectral variability and facilitate the framework to be an efficient model. Furthermore, in the metric learning module, we propose an innovative metric loss function to enhance the discrimination of the embedding features, i.e., interclass far apart and intraclass close. In the experiments, we demonstrate that the classification capacity of the proposed SGML can surpass the comparators on three benchmark data sets.
We show how to augment any convolutional network with an attention-based global map to achieve non-local reasoning. We replace the final average pooling by an attention-based aggregation layer akin to a single transformer block, that weights how the patches are involved in the classification decision. We plug this learned aggregation layer with a simplistic patch-based convolutional network parametrized by 2 parameters (width and depth). In contrast with a pyramidal design, this architecture family maintains the input patch resolution across all the layers. It yields surprisingly competitive trade-offs between accuracy and complexity, in particular in terms of memory consumption, as shown by our experiments on various computer vision tasks: object classification, image segmentation and detection.
Despite their impressive performance on image classification tasks, deep networks have a hard time generalizing to unforeseen corruptions of their data. To fix this vulnerability, prior works have built complex data augmentation strategies, combining multiple methods to enrich the training data. However, introducing intricate design choices or heuristics makes it hard to understand which elements of these methods are indeed crucial for improving robustness. In this work, we take a step back and follow a principled approach to achieve robustness to common corruptions. We propose PRIME, a general data augmentation scheme that relies on simple yet rich families of max-entropy image transformations. PRIME outperforms the prior art in terms of corruption robustness, while its simplicity and plug-and-play nature enable combination with other methods to further boost their robustness. We analyze PRIME to shed light on the importance of the mixing strategy on synthesizing corrupted images, and to reveal the robustness-accuracy trade-offs arising in the context of common corruptions. Finally, we show that the computational efficiency of our method allows it to be easily used in both on-line and off-line data augmentation schemes.
The most recent year has witnessed the success of applying the Vision Transformer (ViT) for image classification. However, there are still evidences indicating that ViT often suffers following two aspects, i) the high computation and the memory burden from applying the multiple Transformer layers for pre-training on a large-scale dataset, ii) the over-fitting when training on small datasets from scratch. To address these problems, a novel method, namely, Vision Reservoir computing (ViR), is proposed here for image classification, as a parallel to ViT. By splitting each image into a sequence of tokens with fixed length, the ViR constructs a pure reservoir with a nearly fully connected topology to replace the Transformer module in ViT. Two kinds of deep ViR models are subsequently proposed to enhance the network performance. Comparative experiments between the ViR and the ViT are carried out on several image classification benchmarks. Without any pre-training process, the ViR outperforms the ViT in terms of both model and computational complexity. Specifically, the number of parameters of the ViR is about 15% even 5% of the ViT, and the memory footprint is about 20% to 40% of the ViT. The superiority of the ViR performance is explained by Small-World characteristics, Lyapunov exponents, and memory capacity.
Recently, the Vision Transformer (ViT), which applied the transformer structure to the image classification task, has outperformed convolutional neural networks. However, the high performance of the ViT results from pre-training using a large-size dataset such as JFT-300M, and its dependence on a large dataset is interpreted as due to low locality inductive bias. This paper proposes Shifted Patch Tokenization (SPT) and Locality Self-Attention (LSA), which effectively solve the lack of locality inductive bias and enable it to learn from scratch even on small-size datasets. Moreover, SPT and LSA are generic and effective add-on modules that are easily applicable to various ViTs. Experimental results show that when both SPT and LSA were applied to the ViTs, the performance improved by an average of 2.96% in Tiny-ImageNet, which is a representative small-size dataset. Especially, Swin Transformer achieved an overwhelming performance improvement of 4.08% thanks to the proposed SPT and LSA.
Efficient and adaptive computer vision systems have been proposed to make computer vision tasks, such as image classification and object detection, optimized for embedded or mobile devices. These solutions, quite recent in their origin, focus on optimizing the model (a deep neural network, DNN) or the system by designing an adaptive system with approximation knobs. In spite of several recent efforts, we show that existing solutions suffer from two major drawbacks. First, the system does not consider energy consumption of the models while making a decision on which model to run. Second, the evaluation does not consider the practical scenario of contention on the device, due to other co-resident workloads. In this work, we propose an efficient and adaptive video object detection system, Virtuoso, which is jointly optimized for accuracy, energy efficiency, and latency. Underlying Virtuoso is a multi-branch execution kernel that is capable of running at different operating points in the accuracy-energy-latency axes, and a lightweight runtime scheduler to select the best fit execution branch to satisfy the user requirement. To fairly compare with Virtuoso, we benchmark 15 state-of-the-art or widely used protocols, including Faster R-CNN (FRCNN), YOLO v3, SSD, EfficientDet, SELSA, MEGA, REPP, FastAdapt, and our in-house adaptive variants of FRCNN+, YOLO+, SSD+, and EfficientDet+ (our variants have enhanced efficiency for mobiles). With this comprehensive benchmark, Virtuoso has shown superiority to all the above protocols, leading the accuracy frontier at every efficiency level on NVIDIA Jetson mobile GPUs. Specifically, Virtuoso has achieved an accuracy of 63.9%, which is more than 10% higher than some of the popular object detection models, FRCNN at 51.1%, and YOLO at 49.5%.
Vision Transformers (ViT) serve as powerful vision models. Unlike convolutional neural networks, which dominated vision research in previous years, vision transformers enjoy the ability to capture long-range dependencies in the data. Nonetheless, an integral part of any transformer architecture, the self-attention mechanism, suffers from high latency and inefficient memory utilization, making it less suitable for high-resolution input images. To alleviate these shortcomings, hierarchical vision models locally employ self-attention on non-interleaving windows. This relaxation reduces the complexity to be linear in the input size; however, it limits the cross-window interaction, hurting the model performance. In this paper, we propose a new shift-invariant local attention layer, called query and attend (QnA), that aggregates the input locally in an overlapping manner, much like convolutions. The key idea behind QnA is to introduce learned queries, which allow fast and efficient implementation. We verify the effectiveness of our layer by incorporating it into a hierarchical vision transformer model. We show improvements in speed and memory complexity while achieving comparable accuracy with state-of-the-art models. Finally, our layer scales especially well with window size, requiring up-to x10 less memory while being up-to x5 faster than existing methods. The code is publicly available at \url{https://github.com/moabarar/qna}.
Neural ordinary differential equations (NODE) have been proposed as a continuous depth generalization to popular deep learning models such as Residual networks (ResNets). They provide parameter efficiency and automate the model selection process in deep learning models to some extent. However, they lack the much-required uncertainty modelling and robustness capabilities which are crucial for their use in several real-world applications such as autonomous driving and healthcare. We propose a novel and unique approach to model uncertainty in NODE by considering a distribution over the end-time $T$ of the ODE solver. The proposed approach, latent time NODE (LT-NODE), treats $T$ as a latent variable and apply Bayesian learning to obtain a posterior distribution over $T$ from the data. In particular, we use variational inference to learn an approximate posterior and the model parameters. Prediction is done by considering the NODE representations from different samples of the posterior and can be done efficiently using a single forward pass. As $T$ implicitly defines the depth of a NODE, posterior distribution over $T$ would also help in model selection in NODE. We also propose, adaptive latent time NODE (ALT-NODE), which allow each data point to have a distinct posterior distribution over end-times. ALT-NODE uses amortized variational inference to learn an approximate posterior using inference networks. We demonstrate the effectiveness of the proposed approaches in modelling uncertainty and robustness through experiments on synthetic and several real-world image classification data.