Current state-of-the-art approaches for image captioning typically adopt an autoregressive manner, i.e., generating descriptions word by word, which suffers from slow decoding issue and becomes a bottleneck in real-time applications. Non-autoregressive image captioning with continuous iterative refinement, which eliminates the sequential dependence in a sentence generation, can achieve comparable performance to the autoregressive counterparts with a considerable acceleration. Nevertheless, based on a well-designed experiment, we empirically proved that iteration times can be effectively reduced when providing sufficient prior knowledge for the language decoder. Towards that end, we propose a novel two-stage framework, referred to as Semi-Autoregressive Image Captioning (SAIC), to make a better trade-off between performance and speed. The proposed SAIC model maintains autoregressive property in global but relieves it in local. Specifically, SAIC model first jumpily generates an intermittent sequence in an autoregressive manner, that is, it predicts the first word in every word group in order. Then, with the help of the partially deterministic prior information and image features, SAIC model non-autoregressively fills all the skipped words with one iteration. Experimental results on the MS COCO benchmark demonstrate that our SAIC model outperforms the preceding non-autoregressive image captioning models while obtaining a competitive inference speedup. Code is available at https://github.com/feizc/SAIC.
The visual system processes a scene using a sequence of selective glimpses, each driven by spatial and object-based attention. These glimpses reflect what is relevant to the ongoing task and are selected through recurrent processing and recognition of the objects in the scene. In contrast, most models treat attention selection and recognition as separate stages in a feedforward process. Here we show that using capsule networks to create an object-centric hidden representation in an encoder-decoder model with iterative glimpse attention yields effective integration of attention and recognition. We evaluate our model on three multi-object recognition tasks; highly overlapping digits, digits among distracting clutter and house numbers, and show that it learns to effectively move its glimpse window, recognize and reconstruct the objects, all with only the classification as supervision. Our work takes a step toward a general architecture for how to integrate recurrent object-centric representation into the planning of attentional glimpses.
Exploiting past 3D LiDAR scans to predict future point clouds is a promising method for autonomous mobile systems to realize foresighted state estimation, collision avoidance, and planning. In this paper, we address the problem of predicting future 3D LiDAR point clouds given a sequence of past LiDAR scans. Estimating the future scene on the sensor level does not require any preceding steps as in localization or tracking systems and can be trained self-supervised. We propose an end-to-end approach that exploits a 2D range image representation of each 3D LiDAR scan and concatenates a sequence of range images to obtain a 3D tensor. Based on such tensors, we develop an encoder-decoder architecture using 3D convolutions to jointly aggregate spatial and temporal information of the scene and to predict the future 3D point clouds. We evaluate our method on multiple datasets and the experimental results suggest that our method outperforms existing point cloud prediction architectures and generalizes well to new, unseen environments without additional fine-tuning. Our method operates online and is faster than the common LiDAR frame rate of 10 Hz.
Transformers are transforming the landscape of computer vision, especially for recognition tasks. Detection transformers are the first fully end-to-end learning systems for object detection, while vision transformers are the first fully transformer-based architecture for image classification. In this paper, we integrate Vision and Detection Transformers (ViDT) to build an effective and efficient object detector. ViDT introduces a reconfigured attention module to extend the recent Swin Transformer to be a standalone object detector, followed by a computationally efficient transformer decoder that exploits multi-scale features and auxiliary techniques essential to boost the detection performance without much increase in computational load. Extensive evaluation results on the Microsoft COCO benchmark dataset demonstrate that ViDT obtains the best AP and latency trade-off among existing fully transformer-based object detectors, and achieves 49.2AP owing to its high scalability for large models. We will release the code and trained models at https://github.com/naver-ai/vidt
Accurate and automated super-resolution image synthesis is highly desired since it has the great potential to circumvent the need for acquiring high-cost medical scans and a time-consuming preprocessing pipeline of neuroimaging data. However, existing deep learning frameworks are solely designed to predict high-resolution (HR) image from a low-resolution (LR) one, which limits their generalization ability to brain graphs (i.e., connectomes). A small body of works has focused on superresolving brain graphs where the goal is to predict a HR graph from a single LR graph. Although promising, existing works mainly focus on superresolving graphs belonging to the same domain (e.g., functional), overlooking the domain fracture existing between multimodal brain data distributions (e.g., morphological and structural). To this aim, we propose a novel inter-domain adaptation framework namely, Learn to SuperResolve Brain Graphs with Knowledge Distillation Network (L2S-KDnet), which adopts a teacher-student paradigm to superresolve brain graphs. Our teacher network is a graph encoder-decoder that firstly learns the LR brain graph embeddings, and secondly learns how to align the resulting latent representations to the HR ground truth data distribution using an adversarial regularization. Ultimately, it decodes the HR graphs from the aligned embeddings. Next, our student network learns the knowledge of the aligned brain graphs as well as the topological structure of the predicted HR graphs transferred from the teacher. We further leverage the decoder of the teacher to optimize the student network. L2S-KDnet presents the first TS architecture tailored for brain graph super-resolution synthesis that is based on inter-domain alignment. Our experimental results demonstrate substantial performance gains over benchmark methods.
Few-shot segmentation is a challenging dense prediction task, which entails segmenting a novel query image given only a small annotated support set. The key problem is thus to design a method that aggregates detailed information from the support set, while being robust to large variations in appearance and context. To this end, we propose a few-shot segmentation method based on dense Gaussian process (GP) regression. Given the support set, our dense GP learns the mapping from local deep image features to mask values, capable of capturing complex appearance distributions. Furthermore, it provides a principled means of capturing uncertainty, which serves as another powerful cue for the final segmentation, obtained by a CNN decoder. Instead of a one-dimensional mask output, we further exploit the end-to-end learning capabilities of our approach to learn a high-dimensional output space for the GP. Our approach sets a new state-of-the-art on the PASCAL-5$^i$ and COCO-20$^i$ benchmarks, achieving an absolute gain of $+8.4$ mIoU in the COCO-20$^i$ 5-shot setting. Furthermore, the segmentation quality of our approach scales gracefully when increasing the support set size, while achieving robust cross-dataset transfer. Code and trained models are available at \url{https://github.com/joakimjohnander/dgpnet}.
We propose an optimized U-Net architecture for a brain tumor segmentation task in the BraTS21 challenge. To find the optimal model architecture and the learning schedule, we have run an extensive ablation study to test: deep supervision loss, Focal loss, decoder attention, drop block, and residual connections. Additionally, we have searched for the optimal depth of the U-Net encoder, number of convolutional channels and post-processing strategy. Our method won the validation phase and took third place in the test phase. We have open-sourced the code to reproduce our BraTS21 submission at the NVIDIA Deep Learning Examples GitHub Repository.
With the advancement of deep models, research work on image captioning has led to a remarkable gain in raw performance over the last decade, along with increasing model complexity and computational cost. However, surprisingly works on compression of deep networks for image captioning task has received little to no attention. For the first time in image captioning research, we provide an extensive comparison of various unstructured weight pruning methods on three different popular image captioning architectures, namely Soft-Attention, Up-Down and Object Relation Transformer. Following this, we propose a novel end-to-end weight pruning method that performs gradual sparsification based on weight sensitivity to the training loss. The pruning schemes are then extended with encoder pruning, where we show that conducting both decoder pruning and training simultaneously prior to the encoder pruning provides good overall performance. Empirically, we show that an 80% to 95% sparse network (up to 75% reduction in model size) can either match or outperform its dense counterpart. The code and pre-trained models for Up-Down and Object Relation Transformer that are capable of achieving CIDEr scores >120 on the MS-COCO dataset but with only 8.7 MB and 14.5 MB in model size (size reduction of 96% and 94% respectively against dense versions) are publicly available at https://github.com/jiahuei/sparse-image-captioning.
Most existing human matting algorithms tried to separate pure human-only foreground from the background. In this paper, we propose a Virtual Multi-modality Foreground Matting (VMFM) method to learn human-object interactive foreground (human and objects interacted with him or her) from a raw RGB image. The VMFM method requires no additional inputs, e.g. trimap or known background. We reformulate foreground matting as a self-supervised multi-modality problem: factor each input image into estimated depth map, segmentation mask, and interaction heatmap using three auto-encoders. In order to fully utilize the characteristics of each modality, we first train a dual encoder-to-decoder network to estimate the same alpha matte. Then we introduce a self-supervised method: Complementary Learning(CL) to predict deviation probability map and exchange reliable gradients across modalities without label. We conducted extensive experiments to analyze the effectiveness of each modality and the significance of different components in complementary learning. We demonstrate that our model outperforms the state-of-the-art methods.
Convolutional Neural Networks (CNNs) have demonstrated state-of-the-art performance in medical image segmentation tasks. A common feature in most top-performing CNNs is an encoder-decoder architecture inspired by the U-Net. For multi-region brain tumor segmentation, 3D U-Net architecture and its variants provide the most competitive segmentation performances. In this work, we propose an interesting extension of the standard 3D U-Net architecture, specialized for brain tumor segmentation. The proposed network, called E1D3 U-Net, is a one-encoder, three-decoder fully-convolutional neural network architecture where each decoder segments one of the hierarchical regions of interest: whole tumor, tumor core, and enhancing core. On the BraTS 2018 validation (unseen) dataset, E1D3 U-Net demonstrates single-prediction performance comparable with most state-of-the-art networks in brain tumor segmentation, with reasonable computational requirements and without ensembling. As a submission to the RSNA-ASNR-MICCAI BraTS 2021 challenge, we also evaluate our proposal on the BraTS 2021 dataset. E1D3 U-Net showcases the flexibility in the standard 3D U-Net architecture which we exploit for the task of brain tumor segmentation.
Local differential privacy (LDP) can be adopted to anonymize richer user data attributes that will be input to sophisticated machine learning (ML) tasks. However, today's LDP approaches are largely task-agnostic and often lead to severe performance loss -- they simply inject noise to all data attributes according to a given privacy budget, regardless of what features are most relevant for the ultimate task. In this paper, we address how to significantly improve the ultimate task performance with multi-dimensional user data by considering a task-aware privacy preservation problem. The key idea is to use an encoder-decoder framework to learn (and anonymize) a task-relevant latent representation of user data. We obtain an analytical near-optimal solution for the linear setting with mean-squared error (MSE) task loss. We also provide an approximate solution through a gradient-based learning algorithm for general nonlinear cases. Extensive experiments demonstrate that our task-aware approach significantly improves ultimate task accuracy compared to standard benchmark LDP approaches with the same level of privacy guarantee.
Existing deep reinforcement learning (DRL) based methods for solving the capacitated vehicle routing problem (CVRP) intrinsically cope with homogeneous vehicle fleet, in which the fleet is assumed as repetitions of a single vehicle. Hence, their key to construct a solution solely lies in the selection of the next node (customer) to visit excluding the selection of vehicle. However, vehicles in real-world scenarios are likely to be heterogeneous with different characteristics that affect their capacity (or travel speed), rendering existing DRL methods less effective. In this paper, we tackle heterogeneous CVRP (HCVRP), where vehicles are mainly characterized by different capacities. We consider both min-max and min-sum objectives for HCVRP, which aim to minimize the longest or total travel time of the vehicle(s) in the fleet. To solve those problems, we propose a DRL method based on the attention mechanism with a vehicle selection decoder accounting for the heterogeneous fleet constraint and a node selection decoder accounting for the route construction, which learns to construct a solution by automatically selecting both a vehicle and a node for this vehicle at each step. Experimental results based on randomly generated instances show that, with desirable generalization to various problem sizes, our method outperforms the state-of-the-art DRL method and most of the conventional heuristics, and also delivers competitive performance against the state-of-the-art heuristic method, i.e., SISR. Additionally, the results of extended experiments demonstrate that our method is also able to solve CVRPLib instances with satisfactory performance.
Polyps represent an early sign of the development of Colorectal Cancer. The standard procedure for their detection consists of colonoscopic examination of the gastrointestinal tract. However, the wide range of polyp shapes and visual appearances, as well as the reduced quality of this image modality, turn their automatic identification and segmentation with computational tools into a challenging computer vision task. In this work, we present a new strategy for the delineation of gastrointestinal polyps from endoscopic images based on a direct extension of common encoder-decoder networks for semantic segmentation. In our approach, two pretrained encoder-decoder networks are sequentially stacked: the second network takes as input the concatenation of the original frame and the initial prediction generated by the first network, which acts as an attention mechanism enabling the second network to focus on interesting areas within the image, thereby improving the quality of its predictions. Quantitative evaluation carried out on several polyp segmentation databases shows that double encoder-decoder networks clearly outperform their single encoder-decoder counterparts in all cases. In addition, our best double encoder-decoder combination attains excellent segmentation accuracy and reaches state-of-the-art performance results in all the considered datasets, with a remarkable boost of accuracy on images extracted from datasets not used for training.
Pre-training (PT) and back-translation (BT) are two simple and powerful methods to utilize monolingual data for improving the model performance of neural machine translation (NMT). This paper takes the first step to investigate the complementarity between PT and BT. We introduce two probing tasks for PT and BT respectively and find that PT mainly contributes to the encoder module while BT brings more benefits to the decoder. Experimental results show that PT and BT are nicely complementary to each other, establishing state-of-the-art performances on the WMT16 English-Romanian and English-Russian benchmarks. Through extensive analyses on sentence originality and word frequency, we also demonstrate that combining Tagged BT with PT is more helpful to their complementarity, leading to better translation quality. Source code is freely available at https://github.com/SunbowLiu/PTvsBT.
In recent years, transformer structures have been widely applied in image captioning with impressive performance. For good captioning results, the geometry and position relations of different visual objects are often thought of as crucial information. Aiming to further promote image captioning by transformers, this paper proposes an improved Geometry Attention Transformer (GAT) model. In order to further leverage geometric information, two novel geometry-aware architectures are designed respectively for the encoder and decoder in our GAT. Besides, this model includes the two work modules: 1) a geometry gate-controlled self-attention refiner, for explicitly incorporating relative spatial information into image region representations in encoding steps, and 2) a group of position-LSTMs, for precisely informing the decoder of relative word position in generating caption texts. The experiment comparisons on the datasets MS COCO and Flickr30K show that our GAT is efficient, and it could often outperform current state-of-the-art image captioning models.
Temporal expressions in text play a significant role in language understanding and correctly identifying them is fundamental to various retrieval and natural language processing systems. Previous works have slowly shifted from rule-based to neural architectures, capable of tagging expressions with higher accuracy. However, neural models can not yet distinguish between different expression types at the same level as their rule-based counterparts. In this work, we aim to identify the most suitable transformer architecture for joint temporal tagging and type classification, as well as, investigating the effect of semi-supervised training on the performance of these systems. Based on our study of token classification variants and encoder-decoder architectures, we present a transformer encoder-decoder model using the RoBERTa language model as our best performing system. By supplementing training resources with weakly labeled data from rule-based systems, our model surpasses previous works in temporal tagging and type classification, especially on rare classes. Our code and pre-trained experiments are available at: https://github.com/satya77/Transformer_Temporal_Tagger
The Transformer and its variants have been proven to be efficient sequence learners in many different domains. Despite their staggering success, a critical issue has been the enormous number of parameters that must be trained (ranging from $10^7$ to $10^{11}$) along with the quadratic complexity of dot-product attention. In this work, we investigate the problem of approximating the two central components of the Transformer -- multi-head self-attention and point-wise feed-forward transformation, with reduced parameter space and computational complexity. We build upon recent developments in analyzing deep neural networks as numerical solvers of ordinary differential equations. Taking advantage of an analogy between Transformer stages and the evolution of a dynamical system of multiple interacting particles, we formulate a temporal evolution scheme, TransEvolve, to bypass costly dot-product attention over multiple stacked layers. We perform exhaustive experiments with TransEvolve on well-known encoder-decoder as well as encoder-only tasks. We observe that the degree of approximation (or inversely, the degree of parameter reduction) has different effects on the performance, depending on the task. While in the encoder-decoder regime, TransEvolve delivers performances comparable to the original Transformer, in encoder-only tasks it consistently outperforms Transformer along with several subsequent variants.
Most of the existing neural video compression methods adopt the predictive coding framework, which first generates the predicted frame and then encodes its residue with the current frame. However, as for compression ratio, predictive coding is only a sub-optimal solution as it uses simple subtraction operation to remove the redundancy across frames. In this paper, we propose a deep contextual video compression framework to enable a paradigm shift from predictive coding to conditional coding. In particular, we try to answer the following questions: how to define, use, and learn condition under a deep video compression framework. To tap the potential of conditional coding, we propose using feature domain context as condition. This enables us to leverage the high dimension context to carry rich information to both the encoder and the decoder, which helps reconstruct the high-frequency contents for higher video quality. Our framework is also extensible, in which the condition can be flexibly designed. Experiments show that our method can significantly outperform the previous state-of-the-art (SOTA) deep video compression methods. When compared with x265 using veryslow preset, we can achieve 26.0% bitrate saving for 1080P standard test videos.
With the COVID-19 global pandemic, computerassisted diagnoses of medical images have gained a lot of attention, and robust methods of Semantic Segmentation of Computed Tomography (CT) turned highly desirable. Semantic Segmentation of CT is one of many research fields of automatic detection of Covid-19 and was widely explored since the Covid19 outbreak. In the robotic field, Semantic Segmentation of organs and CTs are widely used in robots developed for surgery tasks. As new methods and new datasets are proposed quickly, it becomes apparent the necessity of providing an extensive evaluation of those methods. To provide a standardized comparison of different architectures across multiple recently proposed datasets, we propose in this paper an extensive benchmark of multiple encoders and decoders with a total of 120 architectures evaluated in five datasets, with each dataset being validated through a five-fold cross-validation strategy, totaling 3.000 experiments. To the best of our knowledge, this is the largest evaluation in number of encoders, decoders, and datasets proposed in the field of Covid-19 CT segmentation.
Recently, deep clustering methods have gained momentum because of the high representational power of deep neural networks (DNNs) such as autoencoder. The key idea is that representation learning and clustering can reinforce each other: Good representations lead to good clustering while good clustering provides good supervisory signals to representation learning. Critical questions include: 1) How to optimize representation learning and clustering? 2) Should the reconstruction loss of autoencoder be considered always? In this paper, we propose DEKM (for Deep Embedded K-Means) to answer these two questions. Since the embedding space generated by autoencoder may have no obvious cluster structures, we propose to further transform the embedding space to a new space that reveals the cluster-structure information. This is achieved by an orthonormal transformation matrix, which contains the eigenvectors of the within-class scatter matrix of K-means. The eigenvalues indicate the importance of the eigenvectors' contributions to the cluster-structure information in the new space. Our goal is to increase the cluster-structure information. To this end, we discard the decoder and propose a greedy method to optimize the representation. Representation learning and clustering are alternately optimized by DEKM. Experimental results on the real-world datasets demonstrate that DEKM achieves state-of-the-art performance.