Point clouds are often the default choice for many applications as they exhibit more flexibility and efficiency than volumetric data. Nevertheless, their unorganized nature -- points are stored in an unordered way -- makes them less suited to be processed by deep learning pipelines. In this paper, we propose a method for 3D object completion and classification based on point clouds. We introduce a new way of organizing the extracted features based on their activations, which we name soft pooling. For the decoder stage, we propose regional convolutions, a novel operator aimed at maximizing the global activation entropy. Furthermore, inspired by the local refining procedure in Point Completion Network (PCN), we also propose a patch-deforming operation to simulate deconvolutional operations for point clouds. This paper proves that our regional activation can be incorporated in many point cloud architectures like AtlasNet and PCN, leading to better performance for geometric completion. We evaluate our approach on different 3D tasks such as object completion and classification, achieving state-of-the-art accuracy.
Generating diverse and relevant questions over text is a task with widespread applications. We argue that commonly-used evaluation metrics such as BLEU and METEOR are not suitable for this task due to the inherent diversity of reference questions, and propose a scheme for extending conventional metrics to reflect diversity. We furthermore propose a variational encoder-decoder model for this task. We show through automatic and human evaluation that our variational model improves diversity without loss of quality, and demonstrate how our evaluation scheme reflects this improvement.
We present a method for novel view synthesis from input images that are freely distributed around a scene. Our method does not rely on a regular arrangement of input views, can synthesize images for free camera movement through the scene, and works for general scenes with unconstrained geometric layouts. We calibrate the input images via SfM and erect a coarse geometric scaffold via MVS. This scaffold is used to create a proxy depth map for a novel view of the scene. Based on this depth map, a recurrent encoder-decoder network processes reprojected features from nearby views and synthesizes the new view. Our network does not need to be optimized for a given scene. After training on a dataset, it works in previously unseen environments with no fine-tuning or per-scene optimization. We evaluate the presented approach on challenging real-world datasets, including Tanks and Temples, where we demonstrate successful view synthesis for the first time and substantially outperform prior and concurrent work.
Transformer has achieved competitive performance against state-of-the-art end-to-end models in automatic speech recognition (ASR), and requires significantly less training time than RNN-based models. The original Transformer, with encoder-decoder architecture, is only suitable for offline ASR. It relies on an attention mechanism to learn alignments, and encodes input audio bidirectionally. The high computation cost of Transformer decoding also limits its use in production streaming systems. To make Transformer suitable for streaming ASR, we explore Transducer framework as a streamable way to learn alignments. For audio encoding, we apply unidirectional Transformer with interleaved convolution layers. The interleaved convolution layers are used for modeling future context which is important to performance. To reduce computation cost, we gradually downsample acoustic input, also with the interleaved convolution layers. Moreover, we limit the length of history context in self-attention to maintain constant computation cost for each decoding step. We show that this architecture, named Conv-Transformer Transducer, achieves competitive performance on LibriSpeech dataset (3.6\% WER on test-clean) without external language models. The performance is comparable to previously published streamable Transformer Transducer and strong hybrid streaming ASR systems, and is achieved with smaller look-ahead window (140~ms), fewer parameters and lower frame rate.
Sequential reasoning is a complex human ability, with extensive previous research focusing on gaming AI in a single continuous game, round-based decision makings extending to a sequence of games remain less explored. Counter-Strike: Global Offensive (CS:GO), as a round-based game with abundant expert demonstrations, provides an excellent environment for multi-player round-based sequential reasoning. In this work, we propose a Sequence Reasoner with Round Attribute Encoder and Multi-Task Decoder to interpret the strategies behind the round-based purchasing decisions. We adopt few-shot learning to sample multiple rounds in a match, and modified model agnostic meta-learning algorithm Reptile for the meta-learning loop. We formulate each round as a multi-task sequence generation problem. Our state representations combine action encoder, team encoder, player features, round attribute encoder, and economy encoders to help our agent learn to reason under this specific multi-player round-based scenario. A complete ablation study and comparison with the greedy approach certify the effectiveness of our model. Our research will open doors for interpretable AI for understanding episodic and long-term purchasing strategies beyond the gaming community.
This research approaches the task of handwritten text with attention encoder-decoder networks that are trained on Kazakh and Russian language. We developed a novel deep neural network model based on Fully Gated CNN, supported by Multiple bidirectional GRU and Attention mechanisms to manipulate sophisticated features that achieve 0.045 Character Error Rate (CER), 0.192 Word Error Rate (WER) and 0.253 Sequence Error Rate (SER) for the first test dataset and 0.064 CER, 0.24 WER and 0.361 SER for the second test dataset. Also, we propose fully gated layers by taking the advantage of multiple the output feature from Tahn and input feature, this proposed work achieves better results and We experimented with our model on the Handwritten Kazakh & Russian Database (HKR). Our research is the first work on the HKR dataset and demonstrates state-of-the-art results to most of the other existing models.
Attention-based models have made tremendous progress on end-to-end automatic speech recognition(ASR) recently. However, the conventional transformer-based approaches usually generate the sequence results token by token from left to right, leaving the right-to-left contexts unexploited. In this work, we introduce a bidirectional speech transformer to utilize the different directional contexts simultaneously. Specifically, the outputs of our proposed transformer include a left-to-right target, and a right-to-left target. In inference stage, we use the introduced bidirectional beam search method, which can not only generate left-to-right candidates but also generate right-to-left candidates, and determine the best hypothesis by the score. To demonstrate our proposed speech transformer with a bidirectional decoder(STBD), we conduct extensive experiments on the AISHELL-1 dataset. The results of experiments show that STBD achieves a 3.6\% relative CER reduction(CERR) over the unidirectional speech transformer baseline. Besides, the strongest model in this paper called STBD-Big can achieve 6.64\% CER on the test set, without language model rescoring and any extra data augmentation strategies.
This paper presents a novel end-to-end dynamic time-lapse video generation framework, named DTVNet, to generate diversified time-lapse videos from a single landscape image, which are conditioned on normalized motion vectors. The proposed DTVNet consists of two submodules: \emph{Optical Flow Encoder} (OFE) and \emph{Dynamic Video Generator} (DVG). The OFE maps a sequence of optical flow maps to a \emph{normalized motion vector} that encodes the motion information inside the generated video. The DVG contains motion and content streams that learn from the motion vector and the single image respectively, as well as an encoder and a decoder to learn shared content features and construct video frames with corresponding motion respectively. Specifically, the \emph{motion stream} introduces multiple \emph{adaptive instance normalization} (AdaIN) layers to integrate multi-level motion information that are processed by linear layers. In the testing stage, videos with the same content but various motion information can be generated by different \emph{normalized motion vectors} based on only one input image. We further conduct experiments on Sky Time-lapse dataset, and the results demonstrate the superiority of our approach over the state-of-the-art methods for generating high-quality and dynamic videos, as well as the variety for generating videos with various motion information.
Facial image inpainting is a problem that is widely studied, and in recent years the introduction of Generative Adversarial Networks, has led to improvements in the field. Unfortunately some issues persists, in particular when blending the missing pixels with the visible ones. We address the problem by proposing a Wasserstein GAN combined with a new reverse mask operator, namely Reverse Masking Network (R-MNet), a perceptual adversarial network for image inpainting. The reverse mask operator transfers the reverse masked image to the end of the encoder-decoder network leaving only valid pixels to be inpainted. Additionally, we propose a new loss function computed in feature space to target only valid pixels combined with adversarial training. These then capture data distributions and generate images similar to those in the training data with achieved realism (realistic and coherent) on the output images. We evaluate our method on publicly available dataset, and compare with state-of-the-art methods. We show that our method is able to generalize to high-resolution inpainting task, and further show more realistic outputs that are plausible to the human visual system when compared with the state-of-the-art methods.
Measuring the performance of automatic speech recognition (ASR) systems requires manually transcribed data in order to compute the word error rate (WER), which is often time-consuming and expensive. In this paper, we continue our effort in estimating WER using acoustic, lexical and phonotactic features. Our novel approach to estimate the WER uses a multistream end-to-end architecture. We report results for systems using internal speech decoder features (glass-box), systems without speech decoder features (black-box), and for systems without having access to the ASR system (no-box). The no-box system learns joint acoustic-lexical representation from phoneme recognition results along with MFCC acoustic features to estimate WER. Considering WER per sentence, our no-box system achieves 0.56 Pearson correlation with the reference evaluation and 0.24 root mean square error (RMSE) across 1,400 sentences. The estimated overall WER by e-WER2 is 30.9% for a three hours test set, while the WER computed using the reference transcriptions was 28.5%.
Two new approaches to accent classification and conversion are presented and explored, respectively. The first topic is Chinese accent classification/recognition. The second topic is the use of encoder-decoder models for end-to-end Chinese accent conversion, where the classifier in the first topic is used for the training of the accent converter encoder-decoder model. Experiments using different features and model are performed for accent recognition. These features include MFCCs and spectrograms. The classifier models were TDNN and 1D-CNN. On the MAGICDATA dataset with 5 classes of accents, the TDNN classifier trained on MFCC features achieved a test accuracy of 54% and a test F1 score of 0.54 while the 1D-CNN classifier trained on spectrograms achieve a test accuracy of 62% and a test F1 score of 0.62. A prototype of an end-to-end accent converter model is also presented. The converter model comprises of an encoder and a decoder. The encoder model converts an accented input into an accent-neutral form. The decoder model converts an accent-neutral form to an accented form with the specified accent assigned by the input accent label. The converter prototype preserves the tone and foregoes the details in the output audio. An encoder-decoder structure demonstrates the potential of being an effective accent converter. A proposal for future improvements is also presented to address the issue of lost details in the decoder output.
In this paper, we propose a novel way of addressing text-dependent automatic speaker verification (TD-ASV) by using a shared-encoder with task-specific decoders. An autoregressive predictive coding (APC) encoder is pre-trained in an unsupervised manner using both out-of-domain (LibriSpeech, VoxCeleb) and in-domain (DeepMine) unlabeled datasets to learn generic, high-level feature representation that encapsulates speaker and phonetic content. Two task-specific decoders were trained using labeled datasets to classify speakers (SID) and phrases (PID). Speaker embeddings extracted from the SID decoder were scored using a PLDA. SID and PID systems were fused at the score level. There is a 51.9% relative improvement in minDCF for our system compared to the fully supervised x-vector baseline on the cross-lingual DeepMine dataset. However, the i-vector/HMM method outperformed the proposed APC encoder-decoder system. A fusion of the x-vector/PLDA baseline and the SID/PLDA scores prior to PID fusion further improved performance by 15% indicating complementarity of the proposed approach to the x-vector system. We show that the proposed approach can leverage from large, unlabeled, data-rich domains, and learn speech patterns independent of downstream tasks. Such a system can provide competitive performance in domain-mismatched scenarios where test data is from data-scarce domains.
State-of-the-art self-supervised learning approaches for monocular depth estimation usually suffer from scale ambiguity. They do not generalize well when applied on distance estimation for complex projection models such as in fisheye and omnidirectional cameras. This paper introduces a novel multi-task learning strategy to improve self-supervised monocular distance estimation on fisheye and pinhole camera images. Our contribution to this work is threefold: Firstly, we introduce a novel distance estimation network architecture using a self-attention based encoder coupled with robust semantic feature guidance to the decoder that can be trained in a one-stage fashion. Secondly, we integrate a generalized robust loss function, which improves performance significantly while removing the need for hyperparameter tuning with the reprojection loss. Finally, we reduce the artifacts caused by dynamic objects violating static world assumptions using a semantic masking strategy. We significantly improve upon the RMSE of previous work on fisheye by 25% reduction in RMSE. As there is little work on fisheye cameras, we evaluated the proposed method on KITTI using a pinhole model. We achieved state-of-the-art performance among self-supervised methods without requiring an external scale estimation.
Singing voice conversion aims to convert singer's voice from source to target without changing singing content. Parallel training data is typically required for the training of singing voice conversion system, that is however not practical in real-life applications. Recent encoder-decoder structures, such as variational autoencoding Wasserstein generative adversarial network (VAW-GAN), provide an effective way to learn a mapping through non-parallel training data. In this paper, we propose a singing voice conversion framework that is based on VAW-GAN. We train an encoder to disentangle singer identity and singing prosody (F0 contour) from phonetic content. By conditioning on singer identity and F0, the decoder generates output spectral features with unseen target singer identity, and improves the F0 rendering. Experimental results show that the proposed framework achieves better performance than the baseline frameworks.
Many point-based semantic segmentation methods have been designed for indoor scenarios, but they struggle if they are applied to point clouds that are captured by a LiDAR sensor in an outdoor environment. In order to make these methods more efficient and robust such that they can handle LiDAR data, we introduce the general concept of reformulating 3D point-based operations such that they can operate in the projection space. While we show by means of three point-based methods that the reformulated versions are between 300 and 400 times faster and achieve a higher accuracy, we furthermore demonstrate that the concept of reformulating 3D point-based operations allows to design new architectures that unify the benefits of point-based and image-based methods. As an example, we introduce a network that integrates reformulated 3D point-based operations into a 2D encoder-decoder architecture that fuses the information from different 2D scales. We evaluate the approach on four challenging datasets for semantic LiDAR point cloud segmentation and show that leveraging reformulated 3D point-based operations with 2D image-based operations achieves very good results for all four datasets.
Increasing the visibility of nighttime hazy images is challenging because of uneven illumination from active artificial light sources and haze absorbing/scattering. The absence of large-scale benchmark datasets hampers progress in this area. To address this issue, we propose a novel synthetic method called 3R to simulate nighttime hazy images from daytime clear images, which first reconstructs the scene geometry, then simulates the light rays and object reflectance, and finally renders the haze effects. Based on it, we generate realistic nighttime hazy images by sampling real-world light colors from a prior empirical distribution. Experiments on the synthetic benchmark show that the degrading factors jointly reduce the image quality. To address this issue, we propose an optimal-scale maximum reflectance prior to disentangle the color correction from haze removal and address them sequentially. Besides, we also devise a simple but effective learning-based baseline which has an encoder-decoder structure based on the MobileNet-v2 backbone. Experiment results demonstrate their superiority over state-of-the-art methods in terms of both image quality and runtime. Both the dataset and source code will be available at https://github.com/chaimi2013/3R.
We propose a new approach to Human Activity Evaluation (HAE) in long videos using graph-based multi-task modeling. Previous works in activity evaluation either directly compute a metric using a detected skeleton or use the scene information to regress the activity score. These approaches are insufficient for accurate activity assessment since they only compute an average score over a clip, and do not consider the correlation between the joints and body dynamics. Moreover, they are highly scene-dependent which makes the generalizability of these methods questionable. We propose a novel multi-task framework for HAE that utilizes a Graph Convolutional Network backbone to embed the interconnections between human joints in the features. In this framework, we solve the Human Activity Segmentation (HAS) problem as an auxiliary task to improve activity assessment. The HAS head is powered by an Encoder-Decoder Temporal Convolutional Network to semantically segment long videos into distinct activity classes, whereas, HAE uses a Long-Short-Term-Memory-based architecture. We evaluate our method on the UW-IOM and TUM Kitchen datasets and discuss the success and failure cases in these two datasets.
Retinal images have been widely used by clinicians for early diagnosis of ocular diseases. However, the quality of retinal images is often clinically unsatisfactory due to eye lesions and imperfect imaging process. One of the most challenging quality degradation issues in retinal images is non-uniform which hinders the pathological information and further impairs the diagnosis of ophthalmologists and computer-aided analysis.To address this issue, we propose a non-uniform illumination removal network for retinal image, called NuI-Go, which consists of three Recursive Non-local Encoder-Decoder Residual Blocks (NEDRBs) for enhancing the degraded retinal images in a progressive manner. Each NEDRB contains a feature encoder module that captures the hierarchical feature representations, a non-local context module that models the context information, and a feature decoder module that recovers the details and spatial dimension. Additionally, the symmetric skip-connections between the encoder module and the decoder module provide long-range information compensation and reuse. Extensive experiments demonstrate that the proposed method can effectively remove the non-uniform illumination on retinal images while well preserving the image details and color. We further demonstrate the advantages of the proposed method for improving the accuracy of retinal vessel segmentation.
Increasing demand for on-device Automatic Speech Recognition (ASR) systems has resulted in renewed interests in developing automatic model compression techniques. Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7x, outperforming laborious manual rank-selection approaches. However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER). In this work, we propose an iterative AutoML-based LRF approach that achieves over 5x compression without degrading the WER, thereby advancing the state-of-the-art in ASR compression.
Lipreading is an impressive technique and there has been a definite improvement of accuracy in recent years. However, existing methods for lipreading mainly build on autoregressive (AR) model, which generate target tokens one by one and suffer from high inference latency. To breakthrough this constraint, we propose FastLR, a non-autoregressive (NAR) lipreading model which generates all target tokens simultaneously. NAR lipreading is a challenging task that has many difficulties: 1) the discrepancy of sequence lengths between source and target makes it difficult to estimate the length of the output sequence; 2) the conditionally independent behavior of NAR generation lacks the correlation across time which leads to a poor approximation of target distribution; 3) the feature representation ability of encoder can be weak due to lack of effective alignment mechanism; and 4) the removal of AR language model exacerbates the inherent ambiguity problem of lipreading. Thus, in this paper, we introduce three methods to reduce the gap between FastLR and AR model: 1) to address challenges 1 and 2, we leverage integrate-and-fire (I\&F) module to model the correspondence between source video frames and output text sequence. 2) To tackle challenge 3, we add an auxiliary connectionist temporal classification (CTC) decoder to the top of the encoder and optimize it with extra CTC loss. We also add an auxiliary autoregressive decoder to help the feature extraction of encoder. 3) To overcome challenge 4, we propose a novel Noisy Parallel Decoding (NPD) for I\&F and bring Byte-Pair Encoding (BPE) into lipreading. Our experiments exhibit that FastLR achieves the speedup up to 10.97$\times$ comparing with state-of-the-art lipreading model with slight WER absolute increase of 1.5\% and 5.5\% on GRID and LRS2 lipreading datasets respectively, which demonstrates the effectiveness of our proposed method.