Vision transformers (ViTs) have dominated computer vision in recent years. However, ViTs are computationally expensive and not well suited for mobile devices; this led to the prevalence of convolutional neural network (CNN) and ViT-based hybrid models for mobile vision applications. Recently, Vision GNN (ViG) and CNN hybrid models have also been proposed for mobile vision tasks. However, all of these methods remain slower compared to pure CNN-based models. In this work, we propose Multi-Level Dilated Convolutions to devise a purely CNN-based mobile backbone. Using Multi-Level Dilated Convolutions allows for a larger theoretical receptive field than standard convolutions. Different levels of dilation also allow for interactions between the short-range and long-range features in an image. Experiments show that our proposed model outperforms state-of-the-art (SOTA) mobile CNN, ViT, ViG, and hybrid architectures in terms of accuracy and/or speed on image classification, object detection, instance segmentation, and semantic segmentation. Our fastest model, RapidNet-Ti, achieves 76.3\% top-1 accuracy on ImageNet-1K with 0.9 ms inference latency on an iPhone 13 mini NPU, which is faster and more accurate than MobileNetV2x1.4 (74.7\% top-1 with 1.0 ms latency). Our work shows that pure CNN architectures can beat SOTA hybrid and ViT models in terms of accuracy and speed when designed properly.
Estimating crowd count in densely crowded scenes is an extremely challenging task due to non-uniform scale variations. In this paper, we propose a novel end-to-end cascaded network of CNNs to jointly learn crowd count classification and density map estimation. Classifying crowd count into various groups is tantamount to coarsely estimating the total count in the image thereby incorporating a high-level prior into the density estimation network. This enables the layers in the network to learn globally relevant discriminative features which aid in estimating highly refined density maps with lower count error. The joint training is performed in an end-to-end fashion. Extensive experiments on highly challenging publicly available datasets show that the proposed method achieves lower count error and better quality density maps as compared to the recent state-of-the-art methods.
Autoencoders have been successful in learning meaningful representations from image datasets. However, their performance on text datasets has not been widely studied. Traditional autoencoders tend to learn possibly trivial representations of text documents due to their confounding properties such as high-dimensionality, sparsity and power-law word distributions. In this paper, we propose a novel k-competitive autoencoder, called KATE, for text documents. Due to the competition between the neurons in the hidden layer, each neuron becomes specialized in recognizing specific data patterns, and overall the model can learn meaningful representations of textual data. A comprehensive set of experiments show that KATE can learn better representations than traditional autoencoders including denoising, contractive, variational, and k-sparse autoencoders. Our model also outperforms deep generative models, probabilistic topic models, and even word representation models (e.g., Word2Vec) in terms of several downstream tasks such as document classification, regression, and retrieval.
The main essence of this paper is to investigate the performance of RetinaNet based object detectors on pedestrian detection. Pedestrian detection is an important research topic as it provides a baseline for general object detection and has a great number of practical applications like autonomous car, robotics and Security camera. Though extensive research has made huge progress in pedestrian detection, there are still many issues and open for more research and improvement. Recent deep learning based methods have shown state-of-the-art performance in computer vision tasks such as image classification, object detection, and segmentation. Wider pedestrian detection challenge aims at finding improve solutions for pedestrian detection problem. In this paper, We propose a pedestrian detection system based on RetinaNet. Our solution has scored 0.4061 mAP. The code is available at https://github.com/miltonbd/ECCV_2018_pedestrian_detection_challenege.
Despite their renowned predictive power on i.i.d. data, convolutional neural networks are known to rely more on high-frequency patterns that humans deem superficial than on low-frequency patterns that agree better with intuitions about what constitutes category membership. This paper proposes a method for training robust convolutional networks by penalizing the predictive power of the local representations learned by earlier layers. Intuitively, our networks are forced to discard predictive signals such as color and texture that can be gleaned from local receptive fields and to rely instead on the global structures of the image. Across a battery of synthetic and benchmark domain adaptation tasks, our method confers improved generalization out of the domain. Also, to evaluate cross-domain transfer, we introduce ImageNet-Sketch, a new dataset consisting of sketch-like images, that matches the ImageNet classification validation set in categories and scale.
The state of the art in semantic segmentation is steadily increasing in performance, resulting in more precise and reliable segmentations in many different applications. However, progress is limited by the cost of generating labels for training, which sometimes requires hours of manual labor for a single image. Because of this, semi-supervised methods have been applied to this task, with varying degrees of success. A key challenge is that common augmentations used in semi-supervised classification are less effective for semantic segmentation. We propose a novel data augmentation mechanism called ClassMix, which generates augmentations by mixing unlabelled samples, by leveraging on the network's predictions for respecting object boundaries. We evaluate this augmentation technique on two common semi-supervised semantic segmentation benchmarks, showing that it attains state-of-the-art results. Lastly, we also provide extensive ablation studies comparing different design decisions and training regimes.
Neural Architecture Search (NAS) has proved effective in offering outperforming alternatives to handcrafted neural networks. In this paper we analyse the benefits of NAS for image classification tasks under strict computational constraints. Our aim is to automate the design of highly efficient deep neural networks, capable of offering fast and accurate predictions and that could be deployed on a low-memory, low-power system-on-chip. The task thus becomes a three-party trade-off between accuracy, computational complexity, and memory requirements. To address this concern, we propose Multi-Scale Resource-Aware Neural Architecture Search (MS-RANAS). We employ a one-shot architecture search approach in order to obtain a reduced search cost and we focus on an anytime prediction setting. Through the usage of multiple-scaled features and early classifiers, we achieved state-of-the-art results in terms of accuracy-speed trade-off.
Several automatic approaches for objective music performance assessment (MPA) have been proposed in the past, however, existing systems are not yet capable of reliably predicting ratings with the same accuracy as professional judges. This study investigates contrastive learning as a potential method to improve existing MPA systems. Contrastive learning is a widely used technique in representation learning to learn a structured latent space capable of separately clustering multiple classes. It has been shown to produce state of the art results for image-based classification problems. We introduce a weighted contrastive loss suitable for regression tasks applied to a convolutional neural network and show that contrastive loss results in performance gains in regression tasks for MPA. Our results show that contrastive-based methods are able to match and exceed SoTA performance for MPA regression tasks by creating better class clusters within the latent space of the neural networks.
Deep convolutional classifiers linearly separate image classes and improve accuracy as depth increases. They progressively reduce the spatial dimension whereas the number of channels grows with depth. Spatial variability is therefore transformed into variability along channels. A fundamental challenge is to understand the role of non-linearities together with convolutional filters in this transformation. ReLUs with biases are often interpreted as thresholding operators that improve discrimination through sparsity. This paper demonstrates that it is a different mechanism called phase collapse which eliminates spatial variability while linearly separating classes. We show that collapsing the phases of complex wavelet coefficients is sufficient to reach the classification accuracy of ResNets of similar depths. However, replacing the phase collapses with thresholding operators that enforce sparsity considerably degrades the performance. We explain these numerical results by showing that the iteration of phase collapses progressively improves separation of classes, as opposed to thresholding non-linearities.
Person re-identification has attracted many researchers' attention for its wide application, but it is still a very challenging task because only part of the image information can be used for personnel matching. Most of current methods uses CNN to learn to embeddings that can capture semantic similarity information among data points. Many of the state-of-the-arts methods use complex network structures with multiple branches that fuse multiple features while training or testing, using classification loss, Triplet loss or a combination of the two as loss function. However, the method that using Triplet loss as loss function converges slowly, and the method in which pull features of the same class as close as possible in features space leads to poor feature stability. This paper will combine the ranking motivated structured loss, proposed a new metric learning loss function that make the features of the same class are sparsely distributed into the range of small hyperspheres and the features of different classes are uniformly distributed at a clearly angle. And adopted a new single-branch network structure that only using global feature can also get great performance. The validity of our method is verified on the Market1501 and DukeMTMC-ReID person re-identification datasets. Finally acquires 90.9% rank-1 accuracy and 80.8% mAP on DukeMTMC-reID, 95.3% rank-1 accuracy and 88.7% mAP on Market1501. Codes and models are available in Github.https://github.com/Qidian213/Ranked_Person_ReID.
Modern neural network architectures use structured linear transformations, such as low-rank matrices, sparse matrices, permutations, and the Fourier transform, to improve inference speed and reduce memory usage compared to general linear maps. However, choosing which of the myriad structured transformations to use (and its associated parameterization) is a laborious task that requires trading off speed, space, and accuracy. We consider a different approach: we introduce a family of matrices called kaleidoscope matrices (K-matrices) that provably capture any structured matrix with near-optimal space (parameter) and time (arithmetic operation) complexity. We empirically validate that K-matrices can be automatically learned within end-to-end pipelines to replace hand-crafted procedures, in order to improve model quality. For example, replacing channel shuffles in ShuffleNet improves classification accuracy on ImageNet by up to 5%. K-matrices can also simplify hand-engineered pipelines -- we replace filter bank feature computation in speech data preprocessing with a learnable kaleidoscope layer, resulting in only 0.4% loss in accuracy on the TIMIT speech recognition task. In addition, K-matrices can capture latent structure in models: for a challenging permuted image classification task, a K-matrix based representation of permutations is able to learn the right latent structure and improves accuracy of a downstream convolutional model by over 9%. We provide a practically efficient implementation of our approach, and use K-matrices in a Transformer network to attain 36% faster end-to-end inference speed on a language translation task.
Music information is often conveyed or recorded across multiple data modalities including but not limited to audio, images, text and scores. However, music information retrieval research has almost exclusively focused on single modality recognition, requiring development of separate models for each modality. Some multi-modal works require multiple coexisting modalities given to the model as inputs, constraining the use of these models to the few cases where data from all modalities are available. To the best of our knowledge, no existing model has the ability to take inputs from varying modalities, e.g. images or sounds, and classify them into unified music categories. We explore the use of cross-modal retrieval as a pretext task to learn modality-agnostic representations, which can then be used as inputs to classifiers that are independent of modality. We select instrument classification as an example task for our study as both visual and audio components provide relevant semantic information. We train music instrument classifiers that can take both images or sounds as input, and perform comparably to sound-only or image-only classifiers. Furthermore, we explore the case when there is limited labeled data for a given modality, and the impact in performance by using labeled data from other modalities. We are able to achieve almost 70% of best performing system in a zero-shot setting. We provide a detailed analysis of experimental results to understand the potential and limitations of the approach, and discuss future steps towards modality-agnostic classifiers.
Pre-training & fine-tuning can enhance the transferring efficiency and performance in visual tasks. Recent delta-tuning methods provide more options for visual classification tasks. Despite their success, existing visual delta-tuning art fails to exceed the upper limit of full fine-tuning on challenging tasks like instance segmentation and semantic segmentation. To find a competitive alternative to full fine-tuning, we propose the Multi-cognitive Visual Adapter (Mona) tuning, a novel adapter-based tuning method. First, we introduce multiple vision-friendly filters into the adapter to enhance its ability to process visual signals, while previous methods mainly rely on language-friendly linear filters. Second, we add the scaled normalization layer in the adapter to regulate the distribution of input features for visual filters. To fully demonstrate the practicality and generality of Mona, we conduct experiments on multiple representative visual tasks, including instance segmentation on COCO, semantic segmentation on ADE20K, object detection on Pascal VOC, and image classification on several common datasets. Exciting results illustrate that Mona surpasses full fine-tuning on all these tasks and is the only delta-tuning method outperforming full fine-tuning on instance segmentation and semantic segmentation tasks. For example, Mona achieves a 1% performance gain on the COCO dataset compared to full fine-tuning. Comprehensive results suggest that Mona-tuning is more suitable for retaining and utilizing the capabilities of pre-trained models than full fine-tuning. The code will be released at https://github.com/Leiyi-Hu/mona.
The Generative Adversarial Network (GAN) is a state-of-the-art technique in the field of deep learning. A number of recent papers address the theory and applications of GANs in various fields of image processing. Fewer studies, however, have directly evaluated GAN outputs. Those that have been conducted focused on using classification performance, e.g., Inception Score (IS) and statistical metrics, e.g., Fr\'echet Inception Distance (FID). Here, we consider a fundamental way to evaluate GANs by directly analyzing the images they generate, instead of using them as inputs to other classifiers. We characterize the performance of a GAN as an image generator according to three aspects: 1) Creativity: non-duplication of the real images. 2) Inheritance: generated images should have the same style, which retains key features of the real images. 3) Diversity: generated images are different from each other. A GAN should not generate a few different images repeatedly. Based on the three aspects of ideal GANs, we have designed the Likeness Score (LS) to evaluate GAN performance, and have applied it to evaluate several typical GANs. We compared our proposed measure with two commonly used GAN evaluation methods: IS and FID, and four additional measures. Furthermore, we discuss how these evaluations could help us deepen our understanding of GANs and improve their performance.
DEtection TRansformer (DETR) for object detection reaches competitive performance compared with Faster R-CNN via a transformer encoder-decoder architecture. However, trained with scratch transformers, DETR needs large-scale training data and an extreme long training schedule even on COCO dataset. Inspired by the great success of pre-training transformers in natural language processing, we propose a novel pretext task named random query patch detection in Unsupervised Pre-training DETR (UP-DETR). Specifically, we randomly crop patches from the given image and then feed them as queries to the decoder. The model is pre-trained to detect these query patches from the input image. During the pre-training, we address two critical issues: multi-task learning and multi-query localization. (1) To trade off classification and localization preferences in the pretext task, we find that freezing the CNN backbone is the prerequisite for the success of pre-training transformers. (2) To perform multi-query localization, we develop UP-DETR with multi-query patch detection with attention mask. Besides, UP-DETR also provides a unified perspective for fine-tuning object detection and one-shot detection tasks. In our experiments, UP-DETR significantly boosts the performance of DETR with faster convergence and higher average precision on object detection, one-shot detection and panoptic segmentation. Code and pre-training models: https://github.com/dddzg/up-detr.
By leveraging contrastive learning, clustering, and other pretext tasks, unsupervised methods for learning image representations have reached impressive results on standard benchmarks. The result has been a crowded field - many methods with substantially different implementations yield results that seem nearly identical on popular benchmarks, such as linear evaluation on ImageNet. However, a single result does not tell the whole story. In this paper, we compare methods using performance-based benchmarks such as linear evaluation, nearest neighbor classification, and clustering for several different datasets, demonstrating the lack of a clear front-runner within the current state-of-the-art. In contrast to prior work that performs only supervised vs. unsupervised comparison, we compare several different unsupervised methods against each other. To enrich this comparison, we analyze embeddings with measurements such as uniformity, tolerance, and centered kernel alignment (CKA), and propose two new metrics of our own: nearest neighbor graph similarity and linear prediction overlap. We reveal through our analysis that in isolation, single popular methods should not be treated as though they represent the field as a whole, and that future work ought to consider how to leverage the complimentary nature of these methods. We also leverage CKA to provide a framework to robustly quantify augmentation invariance, and provide a reminder that certain types of invariance will be undesirable for downstream tasks.
Data augmentation (DA) is widely used to improve the generalization of neural networks by enforcing the invariances and symmetries to pre-defined transformations applied to input data. However, a fixed augmentation policy may have different effects on each sample in different training stages but existing approaches cannot adjust the policy to be adaptive to each sample and the training model. In this paper, we propose Model Adaptive Data Augmentation (MADAug) that jointly trains an augmentation policy network to teach the model when to learn what. Unlike previous work, MADAug selects augmentation operators for each input image by a model-adaptive policy varying between training stages, producing a data augmentation curriculum optimized for better generalization. In MADAug, we train the policy through a bi-level optimization scheme, which aims to minimize a validation-set loss of a model trained using the policy-produced data augmentations. We conduct an extensive evaluation of MADAug on multiple image classification tasks and network architectures with thorough comparisons to existing DA approaches. MADAug outperforms or is on par with other baselines and exhibits better fairness: it brings improvement to all classes and more to the difficult ones. Moreover, MADAug learned policy shows better performance when transferred to fine-grained datasets. In addition, the auto-optimized policy in MADAug gradually introduces increasing perturbations and naturally forms an easy-to-hard curriculum.
Graph neural networks (GNNs) have achieved state-of-the-art performance in many graph-based tasks such as node classification and graph classification. However, many recent works have demonstrated that an attacker can mislead GNN models by slightly perturbing the graph structure. Existing attacks to GNNs are either under the less practical threat model where the attacker is assumed to access the GNN model parameters, or under the practical black-box threat model but consider perturbing node features that are shown to be not enough effective. In this paper, we aim to bridge this gap and consider black-box attacks to GNNs with structure perturbation as well as with theoretical guarantees. We propose to address this challenge through bandit techniques. Specifically, we formulate our attack as an online optimization with bandit feedback. This original problem is essentially NP-hard due to the fact that perturbing the graph structure is a binary optimization problem. We then propose an online attack based on bandit optimization which is proven to be {sublinear} to the query number $T$, i.e., $\mathcal{O}(\sqrt{N}T^{3/4})$ where $N$ is the number of nodes in the graph. Finally, we evaluate our proposed attack by conducting experiments over multiple datasets and GNN models. The experimental results on various citation graphs and image graphs show that our attack is both effective and efficient. Source code is available at~\url{https://github.com/Metaoblivion/Bandit_GNN_Attack}
We introduce a cluster-based generative image segmentation framework to encode higher-level representations of visual concepts based on one-shot learning inspired by the Omniglot Challenge. The inferred parameters of each component of a Gaussian Mixture Model (GMM) represent a distinct topological subpart of a visual concept. Sampling new data from these parameters generates augmented subparts to build a more robust prototype for each concept, i.e., the Abstracted Gaussian Prototype (AGP). This framework addresses one-shot classification tasks using a cognitively-inspired similarity metric and addresses one-shot generative tasks through a novel AGP-VAE pipeline employing variational autoencoders (VAEs) to generate new class variants. Results from human judges reveal that the generative pipeline produces novel examples and classes of visual concepts that are broadly indistinguishable from those made by humans. The proposed framework leads to impressive but not state-of-the-art classification accuracy; thus, the contribution is two-fold: 1) the system is uniquely low in theoretical and computational complexity and operates in a completely standalone manner compared while existing approaches draw heavily on pre-training or knowledge engineering; and 2) in contrast with competing neural network models, the AGP approach addresses the importance of breadth of task capability emphasized in the Omniglot challenge (i.e., successful performance on generative tasks). These two points are critical as we advance toward an understanding of how learning/reasoning systems can produce viable, robust, and flexible concepts based on literally nothing more than a single example.
Understanding the flow of information in Deep Neural Networks (DNNs) is a challenging problem that has gain increasing attention over the last few years. While several methods have been proposed to explain network predictions, there have been only a few attempts to compare them from a theoretical perspective. What is more, no exhaustive empirical comparison has been performed in the past. In this work, we analyze four gradient-based attribution methods and formally prove conditions of equivalence and approximation between them. By reformulating two of these methods, we construct a unified framework which enables a direct comparison, as well as an easier implementation. Finally, we propose a novel evaluation metric, called Sensitivity-n and test the gradient-based attribution methods alongside with a simple perturbation-based attribution method on several datasets in the domains of image and text classification, using various network architectures.