Table of contents (ToC) extraction aims to extract headings of different levels in documents to better understand the outline of the contents, which can be widely used for document understanding and information retrieval. Existing works often use hand-crafted features and predefined rule-based functions to detect headings and resolve the hierarchical relationship between headings. Both the benchmark and research based on deep learning are still limited. Accordingly, in this paper, we first introduce a standard dataset, HierDoc, including image samples from 650 documents of scientific papers with their content labels. Then we propose a novel end-to-end model by using the multimodal tree decoder (MTD) for ToC as a benchmark for HierDoc. The MTD model is mainly composed of three parts, namely encoder, classifier, and decoder. The encoder fuses the multimodality features of vision, text, and layout information for each entity of the document. Then the classifier recognizes and selects the heading entities. Next, to parse the hierarchical relationship between the heading entities, a tree-structured decoder is designed. To evaluate the performance, both the metric of tree-edit-distance similarity (TEDS) and F1-Measure are adopted. Finally, our MTD approach achieves an average TEDS of 87.2% and an average F1-Measure of 88.1% on the test set of HierDoc. The code and dataset will be released at: https://github.com/Pengfei-Hu/MTD.
Generating new fonts is a time-consuming and labor-intensive task, especially in a language with a huge amount of characters like Chinese. Various deep learning models have demonstrated the ability to efficiently generate new fonts with a few reference characters of that style, but few models support cross-lingual font generation. This paper presents GAS-NeXt, a novel few-shot cross-lingual font generator based on AGIS-Net and Font Translator GAN, and improve the performance metrics such as Fr\'echet Inception Distance (FID), Structural Similarity Index Measure(SSIM), and Pixel-level Accuracy (pix-acc). Our approaches include replacing the original encoder and decoder with the idea of layer attention and context-aware attention from Font Translator GAN, while utilizing the shape, texture, and local discriminators of AGIS-Net. In our experiment on English-to-Chinese font translation, we observed better results in fonts with distinct local features than conventional Chinese fonts compared to results obtained from Font Translator GAN. We also validate our method on multiple languages and datasets.
Unavailability of parallel corpora for training text style transfer (TST) models is a very challenging yet common scenario. Also, TST models implicitly need to preserve the content while transforming a source sentence into the target style. To tackle these problems, an intermediate representation is often constructed that is devoid of style while still preserving the meaning of the source sentence. In this work, we study the usefulness of Abstract Meaning Representation (AMR) graph as the intermediate style agnostic representation. We posit that semantic notations like AMR are a natural choice for an intermediate representation. Hence, we propose T-STAR: a model comprising of two components, text-to-AMR encoder and a AMR-to-text decoder. We propose several modeling improvements to enhance the style agnosticity of the generated AMR. To the best of our knowledge, T-STAR is the first work that uses AMR as an intermediate representation for TST. With thorough experimental evaluation we show T-STAR significantly outperforms state of the art techniques by achieving on an average 15.2% higher content preservation with negligible loss (3% approx.) in style accuracy. Through detailed human evaluation with 90,000 ratings, we also show that T-STAR has up to 50% lesser hallucinations compared to state of the art TST models.
This paper describes the submission of the RoyalFlush neural machine translation system for the WMT 2022 translation efficiency task. Unlike the commonly used autoregressive translation system, we adopted a two-stage translation paradigm called Hybrid Regression Translation (HRT) to combine the advantages of autoregressive and non-autoregressive translation. Specifically, HRT first autoregressively generates a discontinuous sequence (e.g., make a prediction every $k$ tokens, $k>1$) and then fills in all previously skipped tokens at once in a non-autoregressive manner. Thus, we can easily trade off the translation quality and speed by adjusting $k$. In addition, by integrating other modeling techniques (e.g., sequence-level knowledge distillation and deep-encoder-shallow-decoder layer allocation strategy) and a mass of engineering efforts, HRT improves 80\% inference speed and achieves equivalent translation performance with the same-capacity AT counterpart. Our fastest system reaches 6k+ words/second on the GPU latency setting, estimated to be about 3.1x faster than the last year's winner.
We present an effective method for fusing visual-and-language representations for several question answering tasks including visual question answering and visual entailment. In contrast to prior works that concatenate unimodal representations or use only cross-attention, we compose multimodal representations via channel fusion. By fusing on the channels, the model is able to more effectively align the tokens compared to standard methods. These multimodal representations, which we call compound tokens are generated with cross-attention transformer layers. First, vision tokens are used as queries to retrieve compatible text tokens through cross-attention. We then chain the vision tokens and the queried text tokens along the channel dimension. We call the resulting representations compound tokens. A second group of compound tokens are generated using an analogous process where the text tokens serve as queries to the cross-attention layer. We concatenate all the compound tokens for further processing with multimodal encoder. We demonstrate the effectiveness of compound tokens using an encoder-decoder vision-language model trained end-to-end in the open-vocabulary setting. Compound Tokens achieve highly competitive performance across a range of question answering tasks including GQA, VQA2.0, and SNLI-VE.
Neural Radiance Field (NeRF) has revolutionized free viewpoint rendering tasks and achieved impressive results. However, the efficiency and accuracy problems hinder its wide applications. To address these issues, we propose Geometry-Aware Generalized Neural Radiance Field (GARF) with a geometry-aware dynamic sampling (GADS) strategy to perform real-time novel view rendering and unsupervised depth estimation on unseen scenes without per-scene optimization. Distinct from most existing generalized NeRFs, our framework infers the unseen scenes on both pixel-scale and geometry-scale with only a few input images. More specifically, our method learns common attributes of novel-view synthesis by an encoder-decoder structure and a point-level learnable multi-view feature fusion module which helps avoid occlusion. To preserve scene characteristics in the generalized model, we introduce an unsupervised depth estimation module to derive the coarse geometry, narrow down the ray sampling interval to proximity space of the estimated surface and sample in expectation maximum position, constituting Geometry-Aware Dynamic Sampling strategy (GADS). Moreover, we introduce a Multi-level Semantic Consistency loss (MSC) to assist more informative representation learning. Extensive experiments on indoor and outdoor datasets show that comparing with state-of-the-art generalized NeRF methods, GARF reduces samples by more than 25\%, while improving rendering quality and 3D geometry estimation.
In contrast to fully supervised methods using pixel-wise mask labels, box-supervised instance segmentation takes advantage of simple box annotations, which has recently attracted increasing research attention. This paper presents a novel single-shot instance segmentation approach, namely Box2Mask, which integrates the classical level-set evolution model into deep neural network learning to achieve accurate mask prediction with only bounding box supervision. Specifically, both the input image and its deep features are employed to evolve the level-set curves implicitly, and a local consistency module based on a pixel affinity kernel is used to mine the local context and spatial relations. Two types of single-stage frameworks, i.e., CNN-based and transformer-based frameworks, are developed to empower the level-set evolution for box-supervised instance segmentation, and each framework consists of three essential components: instance-aware decoder, box-level matching assignment and level-set evolution. By minimizing the level-set energy function, the mask map of each instance can be iteratively optimized within its bounding box annotation. The experimental results on five challenging testbeds, covering general scenes, remote sensing, medical and scene text images, demonstrate the outstanding performance of our proposed Box2Mask approach for box-supervised instance segmentation. In particular, with the Swin-Transformer large backbone, our Box2Mask obtains 42.4% mask AP on COCO, which is on par with the recently developed fully mask-supervised methods. The code is available at: https://github.com/LiWentomng/boxlevelset.
Autoregressive language modeling (ALM) have been successfully used in self-supervised pre-training in Natural language processing (NLP). However, this paradigm has not achieved comparable results with other self-supervised approach in computer vision (e.g., contrastive learning, mask image modeling). In this paper, we try to find the reason why autoregressive modeling does not work well on vision tasks. To tackle this problem, we fully analyze the limitation of visual autoregressive methods and proposed a novel stochastic autoregressive image modeling (named SAIM) by the two simple designs. First, we employ stochastic permutation strategy to generate effective and robust image context which is critical for vision tasks. Second, we create a parallel encoder-decoder training process in which the encoder serves a similar role to the standard vision transformer focus on learning the whole contextual information, and meanwhile the decoder predicts the content of the current position, so that the encoder and decoder can reinforce each other. By introducing stochastic prediction and the parallel encoder-decoder, SAIM significantly improve the performance of autoregressive image modeling. Our method achieves the best accuracy (83.9%) on the vanilla ViT-Base model among methods using only ImageNet-1K data. Transfer performance in downstream tasks also show that our model achieves competitive performance.
This paper considers lossless image compression and presents a learned compression system that can achieve state-of-the-art lossless compression performance but uses only 59K parameters, which is more than 30x less than other learned systems proposed recently in the literature. The explored system is based on a learned pixel-by-pixel lossless image compression method, where each pixel's probability distribution parameters are obtained by processing the pixel's causal neighborhood (i.e. previously encoded/decoded pixels) with a simple neural network comprising 59K parameters. This causality causes the decoder to operate sequentially, i.e. the neural network has to be evaluated for each pixel sequentially, which increases decoding time significantly with common GPU software and hardware. To reduce the decoding time, parallel decoding algorithms are proposed and implemented. The obtained lossless image compression system is compared to traditional and learned systems in the literature in terms of compression performance, encoding-decoding times and computational complexity.
In this paper, we address the problem of image splicing localization with a multi-stream network architecture that processes the raw RGB image in parallel with other handcrafted forensic signals. Unlike previous methods that either use only the RGB images or stack several signals in a channel-wise manner, we propose an encoder-decoder architecture that consists of multiple encoder streams. Each stream is fed with either the tampered image or handcrafted signals and processes them separately to capture relevant information from each one independently. Finally, the extracted features from the multiple streams are fused in the bottleneck of the architecture and propagated to the decoder network that generates the output localization map. We experiment with two handcrafted algorithms, i.e., DCT and Splicebuster. Our proposed approach is benchmarked on three public forensics datasets, demonstrating competitive performance against several competing methods and achieving state-of-the-art results, e.g., 0.898 AUC on CASIA.
Diffusion models have shown great promise for image generation, beating GANs in terms of generation diversity, with comparable image quality. However, their application to 3D shapes has been limited to point or voxel representations that can in practice not accurately represent a 3D surface. We propose a diffusion model for neural implicit representations of 3D shapes that operates in the latent space of an auto-decoder. This allows us to generate diverse and high quality 3D surfaces. We additionally show that we can condition our model on images or text to enable image-to-3D generation and text-to-3D generation using CLIP embeddings. Furthermore, adding noise to the latent codes of existing shapes allows us to explore shape variations.
In this paper, we propose a novel multi-modal multi-task encoder-decoder pre-training framework (MMSpeech) for Mandarin automatic speech recognition (ASR), which employs both unlabeled speech and text data. The main difficulty in speech-text joint pre-training comes from the significant difference between speech and text modalities, especially for Mandarin speech and text. Unlike English and other languages with an alphabetic writing system, Mandarin uses an ideographic writing system where character and sound are not tightly mapped to one another. Therefore, we propose to introduce the phoneme modality into pre-training, which can help capture modality-invariant information between Mandarin speech and text. Specifically, we employ a multi-task learning framework including five self-supervised and supervised tasks with speech and text data. For end-to-end pre-training, we introduce self-supervised speech-to-pseudo-codes (S2C) and phoneme-to-text (P2T) tasks utilizing unlabeled speech and text data, where speech-pseudo-codes pairs and phoneme-text pairs are a supplement to the supervised speech-text pairs. To train the encoder to learn better speech representation, we introduce self-supervised masked speech prediction (MSP) and supervised phoneme prediction (PP) tasks to learn to map speech into phonemes. Besides, we directly add the downstream supervised speech-to-text (S2T) task into the pre-training process, which can further improve the pre-training performance and achieve better recognition results even without fine-tuning. Experiments on AISHELL-1 show that our proposed method achieves state-of-the-art performance, with a more than 40% relative improvement compared with other pre-training methods.
Eliminating ghosting artifacts due to moving objects is a challenging problem in high dynamic range (HDR) imaging. In this letter, we present a hybrid model consisting of a convolutional encoder and a Transformer decoder to generate ghost-free HDR images. In the encoder, a context aggregation network and non-local attention block are adopted to optimize multi-scale features and capture both global and local dependencies of multiple low dynamic range (LDR) images. The decoder based on Swin Transformer is utilized to improve the reconstruction capability of the proposed model. Motivated by the phenomenal difference between the presence and absence of artifacts under the field of structure tensor (ST), we integrate the ST information of LDR images as auxiliary inputs of the network and use ST loss to further constrain artifacts. Different from previous approaches, our network is capable of processing an arbitrary number of input LDR images. Qualitative and quantitative experiments demonstrate the effectiveness of the proposed method by comparing it with existing state-of-the-art HDR deghosting models. Codes are available at https://github.com/pandayuanyu/HSTHdr.
This paper presents a Generative RegIon-to-Text transformer, GRiT, for object understanding. The spirit of GRiT is to formulate object understanding as <region, text> pairs, where region locates objects and text describes objects. For example, the text in object detection denotes class names while that in dense captioning refers to descriptive sentences. Specifically, GRiT consists of a visual encoder to extract image features, a foreground object extractor to localize objects, and a text decoder to generate open-set object descriptions. With the same model architecture, GRiT can understand objects via not only simple nouns, but also rich descriptive sentences including object attributes or actions. Experimentally, we apply GRiT to object detection and dense captioning tasks. GRiT achieves 60.4 AP on COCO 2017 test-dev for object detection and 15.5 mAP on Visual Genome for dense captioning. Code is available at https://github.com/JialianW/GRiT
Lesion segmentation requires both speed and accuracy. In this paper, we propose a simple yet efficient network DSNet, which consists of a encoder based on Transformer and a convolutional neural network(CNN)-based distinct pyramid decoder containing three dual-stream attention (DSA) modules. Specifically, the DSA module fuses features from two adjacent levels through the false positive stream attention (FPSA) branch and the false negative stream attention (FNSA) branch to obtain features with diversified contextual information. We compare our method with various state-of-the-art (SOTA) lesion segmentation methods with several public datasets, including CVC-ClinicDB, Kvasir-SEG, and ISIC-2018 Task 1. The experimental results show that our method achieves SOTA performance in terms of mean Dice coefficient (mDice) and mean Intersection over Union (mIoU) with low model complexity and memory consumption.
Deep reinforcement learning (RL) has been shown to be effective in producing approximate solutions to some vehicle routing problems (VRPs), especially when using policies generated by encoder-decoder attention mechanisms. While these techniques have been quite successful for relatively simple problem instances, there are still under-researched and highly complex VRP variants for which no effective RL method has been demonstrated. In this work we focus on one such VRP variant, which contains multiple trucks and multi-leg routing requirements. In these problems, demand is required to move along sequences of nodes, instead of just from a start node to an end node. With the goal of making deep RL a viable strategy for real-world industrial-scale supply chain logistics, we develop new extensions to existing encoder-decoder attention models which allow them to handle multiple trucks and multi-leg routing requirements. Our models have the advantage that they can be trained for a small number of trucks and nodes, and then embedded into a large supply chain to yield solutions for larger numbers of trucks and nodes. We test our approach on a real supply chain environment arising in the operations of Japanese automotive parts manufacturer Aisin Corporation, and find that our algorithm outperforms Aisin's previous best solution.
Most existing methods realize 3D instance segmentation by extending those models used for 3D object detection or 3D semantic segmentation. However, these non-straightforward methods suffer from two drawbacks: 1) Imprecise bounding boxes or unsatisfactory semantic predictions limit the performance of the overall 3D instance segmentation framework. 2) Existing method requires a time-consuming intermediate step of aggregation. To address these issues, this paper proposes a novel end-to-end 3D instance segmentation method based on Superpoint Transformer, named as SPFormer. It groups potential features from point clouds into superpoints, and directly predicts instances through query vectors without relying on the results of object detection or semantic segmentation. The key step in this framework is a novel query decoder with transformers that can capture the instance information through the superpoint cross-attention mechanism and generate the superpoint masks of the instances. Through bipartite matching based on superpoint masks, SPFormer can implement the network training without the intermediate aggregation step, which accelerates the network. Extensive experiments on ScanNetv2 and S3DIS benchmarks verify that our method is concise yet efficient. Notably, SPFormer exceeds compared state-of-the-art methods by 4.3% on ScanNetv2 hidden test set in terms of mAP and keeps fast inference speed (247ms per frame) simultaneously. Code is available at https://github.com/sunjiahao1999/SPFormer.
Unsupervised image registration commonly adopts U-Net style networks to predict dense displacement fields in the full-resolution spatial domain. For high-resolution volumetric image data, this process is however resource-intensive and time-consuming. To tackle this problem, we propose the Fourier-Net, replacing the expansive path in a U-Net style network with a parameter-free model-driven decoder. Specifically, instead of our Fourier-Net learning to output a full-resolution displacement field in the spatial domain, we learn its low-dimensional representation in a band-limited Fourier domain. This representation is then decoded by our devised model-driven decoder (consisting of a zero padding layer and an inverse discrete Fourier transform layer) to the dense, full-resolution displacement field in the spatial domain. These changes allow our unsupervised Fourier-Net to contain fewer parameters and computational operations, resulting in faster inference speeds. Fourier-Net is then evaluated on two public 3D brain datasets against various state-of-the-art approaches. For example, when compared to a recent transformer-based method, named TransMorph, our Fourier-Net, which only uses 2.2\% of its parameters and 6.66\% of the multiply-add operations, achieves a 0.5\% higher Dice score and an 11.48 times faster inference speed. Code is available at \url{https://github.com/xi-jia/Fourier-Net}.
Predictive monitoring is a subfield of process mining that aims to predict how a running case will unfold in the future. One of its main challenges is forecasting the sequence of activities that will occur from a given point in time -- suffix prediction -- . Most approaches to the suffix prediction problem learn to predict the suffix by learning how to predict the next activity only, not learning from the whole suffix during the training phase. This paper proposes a novel architecture based on an encoder-decoder model with an attention mechanism that decouples the representation learning of the prefixes from the inference phase, predicting only the activities of the suffix. During the inference phase, this architecture is extended with a heuristic search algorithm that improves the selection of the activity for each index of the suffix. Our approach has been tested using 12 public event logs against 6 different state-of-the-art proposals, showing that it significantly outperforms these proposals.
Generating controllable and editable human motion sequences is a key challenge in 3D Avatar generation. It has been labor-intensive to generate and animate human motion for a long time until learning-based approaches have been developed and applied recently. However, these approaches are still task-specific or modality-specific\cite {ahuja2019language2pose}\cite{ghosh2021synthesis}\cite{ferreira2021learning}\cite{li2021ai}. In this paper, we propose ``UDE", the first unified driving engine that enables generating human motion sequences from natural language or audio sequences (see Fig.~\ref{fig:teaser}). Specifically, UDE consists of the following key components: 1) a motion quantization module based on VQVAE that represents continuous motion sequence as discrete latent code\cite{van2017neural}, 2) a modality-agnostic transformer encoder\cite{vaswani2017attention} that learns to map modality-aware driving signals to a joint space, and 3) a unified token transformer (GPT-like\cite{radford2019language}) network to predict the quantized latent code index in an auto-regressive manner. 4) a diffusion motion decoder that takes as input the motion tokens and decodes them into motion sequences with high diversity. We evaluate our method on HumanML3D\cite{Guo_2022_CVPR} and AIST++\cite{li2021learn} benchmarks, and the experiment results demonstrate our method achieves state-of-the-art performance. Project website: \url{https://github.com/zixiangzhou916/UDE/