We present ObSuRF, a method which turns a single image of a scene into a 3D model represented as a set of Neural Radiance Fields (NeRFs), with each NeRF corresponding to a different object. A single forward pass of an encoder network outputs a set of latent vectors describing the objects in the scene. These vectors are used independently to condition a NeRF decoder, defining the geometry and appearance of each object. We make learning more computationally efficient by deriving a novel loss, which allows training NeRFs on RGB-D inputs without explicit ray marching. After confirming that the model performs equal or better than state of the art on three 2D image segmentation benchmarks, we apply it to two multi-object 3D datasets: A multiview version of CLEVR, and a novel dataset in which scenes are populated by ShapeNet models. We find that after training ObSuRF on RGB-D views of training scenes, it is capable of not only recovering the 3D geometry of a scene depicted in a single input image, but also to segment it into objects, despite receiving no supervision in that regard.
In this work, we develop an attention convolutional neural network (CNN) to segment brain tumors from Magnetic Resonance Images (MRI). Further, we predict the survival rate using various machine learning methods. We adopt a 3D UNet architecture and integrate channel and spatial attention with the decoder network to perform segmentation. For survival prediction, we extract some novel radiomic features based on geometry, location, the shape of the segmented tumor and combine them with clinical information to estimate the survival duration for each patient. We also perform extensive experiments to show the effect of each feature for overall survival (OS) prediction. The experimental results infer that radiomic features such as histogram, location, and shape of the necrosis region and clinical features like age are the most critical parameters to estimate the OS.
Recent works on voice conversion (VC) focus on preserving the rhythm and the intonation as well as the linguistic content. To preserve these features from the source, we decompose current non-parallel VC systems into two encoders and one decoder. We analyze each module with several experiments and reassemble the best components to propose Assem-VC, a new state-of-the-art any-to-many non-parallel VC system. We also examine that PPG and Cotatron features are speaker-dependent, and attempt to remove speaker identity with adversarial training. Code and audio samples are available at https://github.com/mindslab-ai/assem-vc.
Typical high quality text-to-speech (TTS) systems today use a two-stage architecture, with a spectrum model stage that generates spectral frames and a vocoder stage that generates the actual audio. High-quality spectrum models usually incorporate the encoder-decoder architecture with self-attention or bi-directional long short-term (BLSTM) units. While these models can produce high quality speech, they often incur O($L$) increase in both latency and real-time factor (RTF) with respect to input length $L$. In other words, longer inputs leads to longer delay and slower synthesis speed, limiting its use in real-time applications. In this paper, we propose a multi-rate attention architecture that breaks the latency and RTF bottlenecks by computing a compact representation during encoding and recurrently generating the attention vector in a streaming manner during decoding. The proposed architecture achieves high audio quality (MOS of 4.31 compared to groundtruth 4.48), low latency, and low RTF at the same time. Meanwhile, both latency and RTF of the proposed system stay constant regardless of input lengths, making it ideal for real-time applications.
Conventional works that learn grasping affordance from demonstrations need to explicitly predict grasping configurations, such as gripper approaching angles or grasping preshapes. Classic motion planners could then sample trajectories by using such predicted configurations. In this work, our goal is instead to fill the gap between affordance discovery and affordance-based policy learning by integrating the two objectives in an end-to-end imitation learning framework based on deep neural networks. From a psychological perspective, there is a close association between attention and affordance. Therefore, with an end-to-end neural network, we propose to learn affordance cues as visual attention that serves as a useful indicating signal of how a demonstrator accomplishes tasks, instead of explicitly modeling affordances. To achieve this, we propose a contrastive learning framework that consists of a Siamese encoder and a trajectory decoder. We further introduce a coupled triplet loss to encourage the discovered affordance cues to be more affordance-relevant. Our experimental results demonstrate that our model with the coupled triplet loss achieves the highest grasping success rate in a simulated robot environment. Our project website can be accessed at https://sites.google.com/asu.edu/affordance-aware-imitation/project.
A key challenge in self-supervised video representation learning is how to effectively capture motion information besides context bias. While most existing works implicitly achieve this with video-specific pretext tasks (e.g., predicting clip orders, time arrows, and paces), we develop a method that explicitly decouples motion supervision from context bias through a carefully designed pretext task. Specifically, we take the keyframes and motion vectors in compressed videos (e.g., in H.264 format) as the supervision sources for context and motion, respectively, which can be efficiently extracted at over 500 fps on the CPU. Then we design two pretext tasks that are jointly optimized: a context matching task where a pairwise contrastive loss is cast between video clip and keyframe features; and a motion prediction task where clip features, passed through an encoder-decoder network, are used to estimate motion features in a near future. These two tasks use a shared video backbone and separate MLP heads. Experiments show that our approach improves the quality of the learned video representation over previous works, where we obtain absolute gains of 16.0% and 11.1% in video retrieval recall on UCF101 and HMDB51, respectively. Moreover, we find the motion prediction to be a strong regularization for video networks, where using it as an auxiliary task improves the accuracy of action recognition with a margin of 7.4%~13.8%.
Semantic segmentation has a wide array of applications ranging from medical-image analysis, scene understanding, autonomous driving and robotic navigation. This work deals with medical image segmentation and in particular with accurate polyp detection and segmentation during colonoscopy examinations. Several convolutional neural network architectures have been proposed to effectively deal with this task and with the problem of segmenting objects at different scale input. The basic architecture in image segmentation consists of an encoder and a decoder: the first uses convolutional filters to extract features from the image, the second is responsible for generating the final output. In this work, we compare some variant of the DeepLab architecture obtained by varying the decoder backbone. We compare several decoder architectures, including ResNet, Xception, EfficentNet, MobileNet and we perturb their layers by substituting ReLU activation layers with other functions. The resulting methods are used to create deep ensembles which are shown to be very effective. Our experimental evaluations show that our best ensemble produces good segmentation results by achieving high evaluation scores with a dice coefficient of 0.884, and a mean Intersection over Union (mIoU) of 0.818 for the Kvasir-SEG dataset. To improve reproducibility and research efficiency the MATLAB source code used for this research is available at GitHub: https://github.com/LorisNanni.
Video block compressive sensing has been studied for use in resource constrained scenarios, such as wireless sensor networks, but the approach still suffers from low performance and long reconstruction time. Inspired by classical distributed video coding, we design a lightweight encoder with computationally intensive operations, such as video frame interpolation, performed at the decoder. Straying from recent trends in training end-to-end neural networks, we propose two algorithms that leverage convolutional neural network components to reconstruct video with greatly reduced reconstruction time. At the encoder, we leverage temporal correlation between frames and deploy adaptive techniques based on compressive measurements from previous frames. At the decoder, we exploit temporal correlation by using video frame interpolation and temporal differential pulse code modulation. Simulations show that our two proposed algorithms, VAL-VFI and VAL-IDA-VFI reconstruct higher quality video, achieving state-of-the-art performance.
Robust multi-agent trajectory prediction is essential for the safe control of robotic systems. A major challenge is to efficiently learn a representation that approximates the true joint distribution of contextual, social, and temporal information to enable planning. We propose Latent Variable Sequential Set Transformers which are encoder-decoder architectures that generate scene-consistent multi-agent trajectories. We refer to these architectures as "AutoBots". The encoder is a stack of interleaved temporal and social multi-head self-attention (MHSA) modules which alternately perform equivariant processing across the temporal and social dimensions. The decoder employs learnable seed parameters in combination with temporal and social MHSA modules allowing it to perform inference over the entire future scene in a single forward pass efficiently. AutoBots can produce either the trajectory of one ego-agent or a distribution over the future trajectories for all agents in the scene. For the single-agent prediction case, our model achieves top results on the global nuScenes vehicle motion prediction leaderboard, and produces strong results on the Argoverse vehicle prediction challenge. In the multi-agent setting, we evaluate on the synthetic partition of TrajNet++ dataset to showcase the model's socially-consistent predictions. We also demonstrate our model on general sequences of sets and provide illustrative experiments modelling the sequential structure of the multiple strokes that make up symbols in the Omniglot data. A distinguishing feature of AutoBots is that all models are trainable on a single desktop GPU (1080 Ti) in under 48h.
We propose NeRF-VAE, a 3D scene generative model that incorporates geometric structure via NeRF and differentiable volume rendering. In contrast to NeRF, our model takes into account shared structure across scenes, and is able to infer the structure of a novel scene -- without the need to re-train -- using amortized inference. NeRF-VAE's explicit 3D rendering process further contrasts previous generative models with convolution-based rendering which lacks geometric structure. Our model is a VAE that learns a distribution over radiance fields by conditioning them on a latent scene representation. We show that, once trained, NeRF-VAE is able to infer and render geometrically-consistent scenes from previously unseen 3D environments using very few input images. We further demonstrate that NeRF-VAE generalizes well to out-of-distribution cameras, while convolutional models do not. Finally, we introduce and study an attention-based conditioning mechanism of NeRF-VAE's decoder, which improves model performance.
Tracking multiple objects in videos relies on modeling the spatial-temporal interactions of the objects. In this paper, we propose a solution named TransMOT, which leverages powerful graph transformers to efficiently model the spatial and temporal interactions among the objects. TransMOT effectively models the interactions of a large number of objects by arranging the trajectories of the tracked objects as a set of sparse weighted graphs, and constructing a spatial graph transformer encoder layer, a temporal transformer encoder layer, and a spatial graph transformer decoder layer based on the graphs. TransMOT is not only more computationally efficient than the traditional Transformer, but it also achieves better tracking accuracy. To further improve the tracking speed and accuracy, we propose a cascade association framework to handle low-score detections and long-term occlusions that require large computational resources to model in TransMOT. The proposed method is evaluated on multiple benchmark datasets including MOT15, MOT16, MOT17, and MOT20, and it achieves state-of-the-art performance on all the datasets.
Stream fusion, also known as system combination, is a common technique in automatic speech recognition for traditional hybrid hidden Markov model approaches, yet mostly unexplored for modern deep neural network end-to-end model architectures. Here, we investigate various fusion techniques for the all-attention-based encoder-decoder architecture known as the transformer, striving to achieve optimal fusion by investigating different fusion levels in an example single-microphone setting with fusion of standard magnitude and phase features. We introduce a novel multi-encoder learning method that performs a weighted combination of two encoder-decoder multi-head attention outputs only during training. Employing then only the magnitude feature encoder in inference, we are able to show consistent improvement on Wall Street Journal (WSJ) with language model and on Librispeech, without increase in runtime or parameters. Combining two such multi-encoder trained models by a simple late fusion in inference, we achieve state-of-the-art performance for transformer-based models on WSJ with a significant WER reduction of 19% relative compared to the current benchmark approach.
Video portraits relighting is critical in user-facing human photography, especially for immersive VR/AR experience. Recent advances still fail to recover consistent relit result under dynamic illuminations from monocular RGB stream, suffering from the lack of video consistency supervision. In this paper, we propose a neural approach for real-time, high-quality and coherent video portrait relighting, which jointly models the semantic, temporal and lighting consistency using a new dynamic OLAT dataset. We propose a hybrid structure and lighting disentanglement in an encoder-decoder architecture, which combines a multi-task and adversarial training strategy for semantic-aware consistency modeling. We adopt a temporal modeling scheme via flow-based supervision to encode the conjugated temporal consistency in a cross manner. We also propose a lighting sampling strategy to model the illumination consistency and mutation for natural portrait light manipulation in real-world. Extensive experiments demonstrate the effectiveness of our approach for consistent video portrait light-editing and relighting, even using mobile computing.
Reconstructing 3D shape from 2D sketches has long been an open problem because the sketches only provide very sparse and ambiguous information. In this paper, we use an encoder/decoder architecture for the sketch to mesh translation. When integrated into a user interface that provides camera parameters for the sketches, this enables us to leverage its latent parametrization to represent and refine a 3D mesh so that its projections match the external contours outlined in the sketch. We will show that this approach is easy to deploy, robust to style changes, and effective. Furthermore, it can be used for shape refinement given only single pen strokes. We compare our approach to state-of-the-art methods on sketches -- both hand-drawn and synthesized -- and demonstrate that we outperform them.
Low-dose computed tomography (CT) allows the reduction of radiation risk in clinical applications at the expense of image quality, which deteriorates the diagnosis accuracy of radiologists. In this work, we present a High-Quality Imaging network (HQINet) for the CT image reconstruction from Low-dose computed tomography (CT) acquisitions. HQINet was a convolutional encoder-decoder architecture, where the encoder was used to extract spatial and temporal information from three contiguous slices while the decoder was used to recover the spacial information of the middle slice. We provide experimental results on the real projection data from low-dose CT Image and Projection Data (LDCT-and-Projection-data), demonstrating that the proposed approach yielded a notable improvement of the performance in terms of image quality, with a rise of 5.5dB in terms of peak signal-to-noise ratio (PSNR) and 0.29 in terms of mutual information (MI).
In this article, we present a new iterative MIMO detection algorithm based on inexact alternating direction method of multipliers. Each iteration is considered as a neural network layer with learnable parameters, which are optimized by the stochastic gradient descent algorithm with a training data set of the received vectors and the ground truth transmitted signals. Numerical results show that the proposed algorithm outperforms the existing learnable detection network and it achieves near-optimal performance close to the sphere decoder in the case of a large number of receive antennas.
The most common tools for word-alignment rely on a large amount of parallel sentences, which are then usually processed according to one of the IBM model algorithms. The training data is, however, the same as for machine translation (MT) systems, especially for neural MT (NMT), which itself is able to produce word-alignments using the trained attention heads. This is convenient because word-alignment is theoretically a viable byproduct of any attention-based NMT, which is also able to provide decoder scores for a translated sentence pair. We summarize different approaches on how word-alignment can be extracted from alignment scores and then explore ways in which scores can be extracted from NMT, focusing on inferring the word-alignment scores based on output sentence and token probabilities. We compare this to the extraction of alignment scores from attention. We conclude with aggregating all of the sources of alignment scores into a simple feed-forward network which achieves the best results when combined alignment extractors are used.
The ubiquitous presence of machine learning systems in our lives necessitates research into their vulnerabilities and appropriate countermeasures. In particular, we investigate the effectiveness of adversarial attacks and defenses against automatic speech recognition (ASR) systems. We select two ASR models - a thoroughly studied DeepSpeech model and a more recent Espresso framework Transformer encoder-decoder model. We investigate two threat models: a denial-of-service scenario where fast gradient-sign method (FGSM) or weak projected gradient descent (PGD) attacks are used to degrade the model's word error rate (WER); and a targeted scenario where a more potent imperceptible attack forces the system to recognize a specific phrase. We find that the attack transferability across the investigated ASR systems is limited. To defend the model, we use two preprocessing defenses: randomized smoothing and WaveGAN-based vocoder, and find that they significantly improve the model's adversarial robustness. We show that a WaveGAN vocoder can be a useful countermeasure to adversarial attacks on ASR systems - even when it is jointly attacked with the ASR, the target phrases' word error rate is high.
Designing a speech-to-intent (S2I) agent which maps the users' spoken commands to the agents' desired task actions can be challenging due to the diverse grammatical and lexical preference of different users. As a remedy, we discuss a user-taught S2I system in this paper. The user-taught system learns from scratch from the users' spoken input with action demonstration, which ensure it is fully matched to the users' way of formulating intents and their articulation habits. The main issue is the scarce training data due to the user effort involved. Existing state-of-art approaches in this setting are based on non-negative matrix factorization (NMF) and capsule networks. In this paper we combine the encoder of an end-to-end ASR system with the prior NMF/capsule network-based user-taught decoder, and investigate whether pre-training methodology can reduce training data requirements for the NMF and capsule network. Experimental results show the pre-trained ASR-NMF framework significantly outperforms other models, and also, we discuss limitations of pre-training with different types of command-and-control(C&C) applications.
In this paper, we present a new tracking architecture with an encoder-decoder transformer as the key component. The encoder models the global spatio-temporal feature dependencies between target objects and search regions, while the decoder learns a query embedding to predict the spatial positions of the target objects. Our method casts object tracking as a direct bounding box prediction problem, without using any proposals or predefined anchors. With the encoder-decoder transformer, the prediction of objects just uses a simple fully-convolutional network, which estimates the corners of objects directly. The whole method is end-to-end, does not need any postprocessing steps such as cosine window and bounding box smoothing, thus largely simplifying existing tracking pipelines. The proposed tracker achieves state-of-the-art performance on five challenging short-term and long-term benchmarks, while running at real-time speed, being 6x faster than Siam R-CNN. Code and models are open-sourced at https://github.com/researchmm/Stark.