Monocular depth estimation is scale-ambiguous, and thus requires scale supervision to produce metric predictions. Even so, the resulting models will be geometry-specific, with learned scales that cannot be directly transferred across domains. Because of that, recent works focus instead on relative depth, eschewing scale in favor of improved up-to-scale zero-shot transfer. In this work we introduce ZeroDepth, a novel monocular depth estimation framework capable of predicting metric scale for arbitrary test images from different domains and camera parameters. This is achieved by (i) the use of input-level geometric embeddings that enable the network to learn a scale prior over objects; and (ii) decoupling the encoder and decoder stages, via a variational latent representation that is conditioned on single frame information. We evaluated ZeroDepth targeting both outdoor (KITTI, DDAD, nuScenes) and indoor (NYUv2) benchmarks, and achieved a new state-of-the-art in both settings using the same pre-trained model, outperforming methods that train on in-domain data and require test-time scaling to produce metric estimates.
The combination of audio and vision has long been a topic of interest in the multi-modal community. Recently, a new audio-visual segmentation (AVS) task has been introduced, aiming to locate and segment the sounding objects in a given video. This task demands audio-driven pixel-level scene understanding for the first time, posing significant challenges. In this paper, we propose AVSegFormer, a novel framework for AVS tasks that leverages the transformer architecture. Specifically, we introduce audio queries and learnable queries into the transformer decoder, enabling the network to selectively attend to interested visual features. Besides, we present an audio-visual mixer, which can dynamically adjust visual features by amplifying relevant and suppressing irrelevant spatial channels. Additionally, we devise an intermediate mask loss to enhance the supervision of the decoder, encouraging the network to produce more accurate intermediate predictions. Extensive experiments demonstrate that AVSegFormer achieves state-of-the-art results on the AVS benchmark. The code is available at https://github.com/vvvb-github/AVSegFormer.
Temporal Point Processes (TPPs) serve as the standard mathematical framework for modeling asynchronous event sequences in continuous time. However, classical TPP models are often constrained by strong assumptions, limiting their ability to capture complex real-world event dynamics. To overcome this limitation, researchers have proposed Neural TPPs, which leverage neural network parametrizations to offer more flexible and efficient modeling. While recent studies demonstrate the effectiveness of Neural TPPs, they often lack a unified setup, relying on different baselines, datasets, and experimental configurations. This makes it challenging to identify the key factors driving improvements in predictive accuracy, hindering research progress. To bridge this gap, we present a comprehensive large-scale experimental study that systematically evaluates the predictive accuracy of state-of-the-art neural TPP models. Our study encompasses multiple real-world and synthetic event sequence datasets, following a carefully designed unified setup. We thoroughly investigate the influence of major architectural components such as event encoding, history encoder, and decoder parametrization on both time and mark prediction tasks. Additionally, we delve into the less explored area of probabilistic calibration for neural TPP models. By analyzing our results, we draw insightful conclusions regarding the significance of history size and the impact of architectural components on predictive accuracy. Furthermore, we shed light on the miscalibration of mark distributions in neural TPP models. Our study aims to provide valuable insights into the performance and characteristics of neural TPP models, contributing to a better understanding of their strengths and limitations.
This paper describes a data-driven framework to parse musical sequences into dependency trees, which are hierarchical structures used in music cognition research and music analysis. The parsing involves two steps. First, the input sequence is passed through a transformer encoder to enrich it with contextual information. Then, a classifier filters the graph of all possible dependency arcs to produce the dependency tree. One major benefit of this system is that it can be easily integrated into modern deep-learning pipelines. Moreover, since it does not rely on any particular symbolic grammar, it can consider multiple musical features simultaneously, make use of sequential context information, and produce partial results for noisy inputs. We test our approach on two datasets of musical trees -- time-span trees of monophonic note sequences and harmonic trees of jazz chord sequences -- and show that our approach outperforms previous methods.
Automated medical image segmentation inherently involves a certain degree of uncertainty. One key factor contributing to this uncertainty is the ambiguity that can arise in determining the boundaries of a target region of interest, primarily due to variations in image appearance. On top of this, even among experts in the field, different opinions can emerge regarding the precise definition of specific anatomical structures. This work specifically addresses the modeling of segmentation uncertainty, known as inter-rater uncertainty. Its primary objective is to explore and analyze the variability in segmentation outcomes that can occur when multiple experts in medical imaging interpret and annotate the same images. We introduce a novel Bayesian neural network-based architecture to estimate inter-rater uncertainty in medical image segmentation. Our approach has three key advancements. Firstly, we introduce a one-encoder-multi-decoder architecture specifically tailored for uncertainty estimation, enabling us to capture the rater-specific representation of each expert involved. Secondly, we propose Bayesian modeling for the new architecture, allowing efficient capture of the inter-rater distribution, particularly in scenarios with limited annotations. Lastly, we enhance the rater-specific representation by integrating an attention module into each decoder. This module facilitates focused and refined segmentation results for each rater. We conduct extensive evaluations using synthetic and real-world datasets to validate our technical innovations rigorously. Our method surpasses existing baseline methods in five out of seven diverse tasks on the publicly available \emph{QUBIQ} dataset, considering two evaluation metrics encompassing different uncertainty aspects. Our codes, models, and the new dataset are available through our GitHub repository: https://github.com/HaoWang420/bOEMD-net .
Multimodal Sarcasm Explanation (MuSE) is a new yet challenging task, which aims to generate a natural language sentence for a multimodal social post (an image as well as its caption) to explain why it contains sarcasm. Although the existing pioneer study has achieved great success with the BART backbone, it overlooks the gap between the visual feature space and the decoder semantic space, the object-level metadata of the image, as well as the potential external knowledge. To solve these limitations, in this work, we propose a novel mulTi-source sEmantic grAph-based Multimodal sarcasm explanation scheme, named TEAM. In particular, TEAM extracts the object-level semantic meta-data instead of the traditional global visual features from the input image. Meanwhile, TEAM resorts to ConceptNet to obtain the external related knowledge concepts for the input text and the extracted object meta-data. Thereafter, TEAM introduces a multi-source semantic graph that comprehensively characterize the multi-source (i.e., caption, object meta-data, external knowledge) semantic relations to facilitate the sarcasm reasoning. Extensive experiments on a public released dataset MORE verify the superiority of our model over cutting-edge methods.
GPU decoding significantly accelerates the output of ASR predictions. While GPUs are already being used for online ASR decoding, post-processing and rescoring on GPUs have not been properly investigated yet. Rescoring with available contextual information can considerably improve ASR predictions. Previous studies have proven the viability of lattice rescoring in decoding and biasing language model (LM) weights in offline and online CPU scenarios. In real-time GPU decoding, partial recognition hypotheses are produced without lattice generation, which makes the implementation of biasing more complex. The paper proposes and describes an approach to integrate contextual biasing in real-time GPU decoding while exploiting the standard Kaldi GPU decoder. Besides the biasing of partial ASR predictions, our approach also permits dynamic context switching allowing a flexible rescoring per each speech segment directly on GPU. The code is publicly released and tested with open-sourced test sets.
Segment Anything Model (SAM) has attracted significant attention due to its impressive zero-shot transfer performance and high versatility for numerous vision applications (like image editing with fine-grained control). Many of such applications need to be run on resource-constraint edge devices, like mobile phones. In this work, we aim to make SAM mobile-friendly by replacing the heavyweight image encoder with a lightweight one. A naive way to train such a new SAM as in the original SAM paper leads to unsatisfactory performance, especially when limited training sources are available. We find that this is mainly caused by the coupled optimization of the image encoder and mask decoder, motivated by which we propose decoupled distillation. Concretely, we distill the knowledge from the heavy image encoder (ViT-H in the original SAM) to a lightweight image encoder, which can be automatically compatible with the mask decoder in the original SAM. The training can be completed on a single GPU within less than one day, and the resulting lightweight SAM is termed MobileSAM which is more than 60 times smaller yet performs on par with the original SAM. For inference speed, With a single GPU, MobileSAM runs around 10ms per image: 8ms on the image encoder and 4ms on the mask decoder. With superior performance, our MobileSAM is around 5 times faster than the concurrent FastSAM and 7 times smaller, making it more suitable for mobile applications. Moreover, we show that MobileSAM can run relatively smoothly on CPU. The code for our project is provided at \href{https://github.com/ChaoningZhang/MobileSAM}{\textcolor{red}{MobileSAM}}), with a demo showing that MobileSAM can run relatively smoothly on CPU.
Zero-shot Learning (ZSL) classification categorizes or predicts classes (labels) that are not included in the training set (unseen classes). Recent works proposed different semantic autoencoder (SAE) models where the encoder embeds a visual feature vector space into the semantic space and the decoder reconstructs the original visual feature space. The objective is to learn the embedding by leveraging a source data distribution, which can be applied effectively to a different but related target data distribution. Such embedding-based methods are prone to domain shift problems and are vulnerable to biases. We propose an integral projection-based semantic autoencoder (IP-SAE) where an encoder projects a visual feature space concatenated with the semantic space into a latent representation space. We force the decoder to reconstruct the visual-semantic data space. Due to this constraint, the visual-semantic projection function preserves the discriminatory data included inside the original visual feature space. The enriched projection forces a more precise reconstitution of the visual feature space invariant to the domain manifold. Consequently, the learned projection function is less domain-specific and alleviates the domain shift problem. Our proposed IP-SAE model consolidates a symmetric transformation function for embedding and projection, and thus, it provides transparency for interpreting generative applications in ZSL. Therefore, in addition to outperforming state-of-the-art methods considering four benchmark datasets, our analytical approach allows us to investigate distinct characteristics of generative-based methods in the unique context of zero-shot inference.
The art of mathematical reasoning stands as a fundamental pillar of intellectual progress and is a central catalyst in cultivating human ingenuity. Researchers have recently published a plethora of works centered around the task of solving Math Word Problems (MWP) $-$ a crucial stride towards general AI. These existing models are susceptible to dependency on shallow heuristics and spurious correlations to derive the solution expressions. In order to ameliorate this issue, in this paper, we propose a framework for MWP solvers based on the generation of linguistic variants of the problem text. The approach involves solving each of the variant problems and electing the predicted expression with the majority of the votes. We use DeBERTa (Decoding-enhanced BERT with disentangled attention) as the encoder to leverage its rich textual representations and enhanced mask decoder to construct the solution expressions. Furthermore, we introduce a challenging dataset, $\mathrm{P\small{ARA}\normalsize{MAWPS}}$, consisting of paraphrased, adversarial, and inverse variants of selectively sampled MWPs from the benchmark $\mathrm{M\small{AWPS}}$ dataset. We extensively experiment on this dataset along with other benchmark datasets using some baseline MWP solver models. We show that training on linguistic variants of problem statements and voting on candidate predictions improve the mathematical reasoning and robustness of the model. We make our code and data publicly available.
Knowledge graphs (KGs), containing many entity-relation-entity triples, provide rich information for downstream applications. Although extracting triples from unstructured texts has been widely explored, most of them require a large number of labeled instances. The performance will drop dramatically when only few labeled data are available. To tackle this problem, we propose the Mutually Guided Few-shot learning framework for Relational Triple Extraction (MG-FTE). Specifically, our method consists of an entity-guided relation proto-decoder to classify the relations firstly and a relation-guided entity proto-decoder to extract entities based on the classified relations. To draw the connection between entity and relation, we design a proto-level fusion module to boost the performance of both entity extraction and relation classification. Moreover, a new cross-domain few-shot triple extraction task is introduced. Extensive experiments show that our method outperforms many state-of-the-art methods by 12.6 F1 score on FewRel 1.0 (single-domain) and 20.5 F1 score on FewRel 2.0 (cross-domain).
Recently, foundation models have been introduced demonstrating various tasks in the field of computer vision. These models such as Segment Anything Model (SAM) are generalized models trained using huge datasets. Currently, ongoing research focuses on exploring the effective utilization of these generalized models for specific domains, such as medical imaging. However, in medical imaging, the lack of training samples due to privacy concerns and other factors presents a major challenge for applying these generalized models to medical image segmentation task. To address this issue, the effective fine tuning of these models is crucial to ensure their optimal utilization. In this study, we propose to combine a complementary Convolutional Neural Network (CNN) along with the standard SAM network for medical image segmentation. To reduce the burden of fine tuning large foundation model and implement cost-efficient trainnig scheme, we focus only on fine-tuning the additional CNN network and SAM decoder part. This strategy significantly reduces trainnig time and achieves competitive results on publicly available dataset. The code is available at https://github.com/11yxk/SAM-LST.
Change detection is of fundamental importance when analyzing data streams. Detecting changes both quickly and accurately enables monitoring and prediction systems to react, e.g., by issuing an alarm or by updating a learning algorithm. However, detecting changes is challenging when observations are high-dimensional. In high-dimensional data, change detectors should not only be able to identify when changes happen, but also in which subspace they occur. Ideally, one should also quantify how severe they are. Our approach, ABCD, has these properties. ABCD learns an encoder-decoder model and monitors its accuracy over a window of adaptive size. ABCD derives a change score based on Bernstein's inequality to detect deviations in terms of accuracy, which indicate changes. Our experiments demonstrate that ABCD outperforms its best competitor by up to 20% in F1-score on average. It can also accurately estimate changes' subspace, together with a severity measure that correlates with the ground truth.
We propose an efficient abnormal event detection model based on a lightweight masked auto-encoder (AE) applied at the video frame level. The novelty of the proposed model is threefold. First, we introduce an approach to weight tokens based on motion gradients, thus shifting the focus from the static background scene to the foreground objects. Second, we integrate a teacher decoder and a student decoder into our architecture, leveraging the discrepancy between the outputs given by the two decoders to improve anomaly detection. Third, we generate synthetic abnormal events to augment the training videos, and task the masked AE model to jointly reconstruct the original frames (without anomalies) and the corresponding pixel-level anomaly maps. Our design leads to an efficient and effective model, as demonstrated by the extensive experiments carried out on four benchmarks: Avenue, ShanghaiTech, UBnormal and UCSD Ped2. The empirical results show that our model achieves an excellent trade-off between speed and accuracy, obtaining competitive AUC scores, while processing 1655 FPS. Hence, our model is between 8 and 70 times faster than competing methods. We also conduct an ablation study to justify our design. Our code is freely available at: https://github.com/ristea/aed-mae.
Scene text removal (STR) aims at replacing text strokes in natural scenes with visually coherent backgrounds. Recent STR approaches rely on iterative refinements or explicit text masks, resulting in high complexity and sensitivity to the accuracy of text localization. Moreover, most existing STR methods adopt convolutional architectures while the potential of vision Transformers (ViTs) remains largely unexplored. In this paper, we propose a simple-yet-effective ViT-based text eraser, dubbed ViTEraser. Following a concise encoder-decoder framework, ViTEraser can easily incorporate various ViTs to enhance long-range modeling. Specifically, the encoder hierarchically maps the input image into the hidden space through ViT blocks and patch embedding layers, while the decoder gradually upsamples the hidden features to the text-erased image with ViT blocks and patch splitting layers. As ViTEraser implicitly integrates text localization and inpainting, we propose a novel end-to-end pretraining method, termed SegMIM, which focuses the encoder and decoder on the text box segmentation and masked image modeling tasks, respectively. Experimental results demonstrate that ViTEraser with SegMIM achieves state-of-the-art performance on STR by a substantial margin and exhibits strong generalization ability when extended to other tasks, \textit{e.g.}, tampered scene text detection. Furthermore, we comprehensively explore the architecture, pretraining, and scalability of the ViT-based encoder-decoder for STR, which provides deep insights into the application of ViT to the STR field. Code is available at https://github.com/shannanyinxiang/ViTEraser.
In recent years, the integration of Large Language Models (LLMs) into recommender systems has garnered interest among both practitioners and researchers. Despite this interest, the field is still emerging, and the lack of open-source R&D platforms may impede the exploration of LLM-based recommendations. This paper introduces OpenP5, an open-source platform designed as a resource to facilitate the development, training, and evaluation of LLM-based generative recommender systems for research purposes. The platform is implemented using encoder-decoder LLMs (e.g., T5) and decoder-only LLMs (e.g., Llama-2) across 10 widely recognized public datasets, catering to two fundamental recommendation tasks: sequential and straightforward recommendations. Recognizing the crucial role of item IDs in LLM-based recommendations, we have also incorporated three item indexing methods within the OpenP5 platform: random indexing, sequential indexing and collaborative indexing. Built on the Transformers library, the platform facilitates easy customization of LLM-based recommendations for users. OpenP5 boasts a range of features including extensible data processing, task-centric optimization, comprehensive datasets and checkpoints, efficient acceleration, and standardized evaluations, making it a valuable tool for the implementation and evaluation of LLM-based recommender systems. The open-source code and pre-trained checkpoints for the OpenP5 library are publicly available at https://github.com/agiresearch/OpenP5.
We present a novel typical-to-atypical voice conversion approach (DuTa-VC), which (i) can be trained with nonparallel data (ii) first introduces diffusion probabilistic model (iii) preserves the target speaker identity (iv) is aware of the phoneme duration of the target speaker. DuTa-VC consists of three parts: an encoder transforms the source mel-spectrogram into a duration-modified speaker-independent mel-spectrogram, a decoder performs the reverse diffusion to generate the target mel-spectrogram, and a vocoder is applied to reconstruct the waveform. Objective evaluations conducted on the UASpeech show that DuTa-VC is able to capture severity characteristics of dysarthric speech, reserves speaker identity, and significantly improves dysarthric speech recognition as a data augmentation. Subjective evaluations by two expert speech pathologists validate that DuTa-VC can preserve the severity and type of dysarthria of the target speakers in the synthesized speech.
The Streaming Unmixing and Recognition Transducer (SURT) model was proposed recently as an end-to-end approach for continuous, streaming, multi-talker speech recognition (ASR). Despite impressive results on multi-turn meetings, SURT has notable limitations: (i) it suffers from leakage and omission related errors; (ii) it is computationally expensive, due to which it has not seen adoption in academia; and (iii) it has only been evaluated on synthetic mixtures. In this work, we propose several modifications to the original SURT which are carefully designed to fix the above limitations. In particular, we (i) change the unmixing module to a mask estimator that uses dual-path modeling, (ii) use a streaming zipformer encoder and a stateless decoder for the transducer, (iii) perform mixture simulation using force-aligned subsegments, (iv) pre-train the transducer on single-speaker data, (v) use auxiliary objectives in the form of masking loss and encoder CTC loss, and (vi) perform domain adaptation for far-field recognition. We show that our modifications allow SURT 2.0 to outperform its predecessor in terms of multi-talker ASR results, while being efficient enough to train with academic resources. We conduct our evaluations on 3 publicly available meeting benchmarks -- LibriCSS, AMI, and ICSI, where our best model achieves WERs of 16.9%, 44.6% and 32.2%, respectively, on far-field unsegmented recordings. We release training recipes and pre-trained models: https://sites.google.com/view/surt2.
Predicting sea surface temperature (SST) within the El Ni\~no-Southern Oscillation (ENSO) region has been extensively studied due to its significant influence on global temperature and precipitation patterns. Statistical models such as linear inverse model (LIM), analog forecasting (AF), and recurrent neural network (RNN) have been widely used for ENSO prediction, offering flexibility and relatively low computational expense compared to large dynamic models. However, these models have limitations in capturing spatial patterns in SST variability or relying on linear dynamics. Here we present a modified Convolutional Gated Recurrent Unit (ConvGRU) network for the ENSO region spatio-temporal sequence prediction problem, along with the Ni\~no 3.4 index prediction as a down stream task. The proposed ConvGRU network, with an encoder-decoder sequence-to-sequence structure, takes historical SST maps of the Pacific region as input and generates future SST maps for subsequent months within the ENSO region. To evaluate the performance of the ConvGRU network, we trained and tested it using data from multiple large climate models. The results demonstrate that the ConvGRU network significantly improves the predictability of the Ni\~no 3.4 index compared to LIM, AF, and RNN. This improvement is evidenced by extended useful prediction range, higher Pearson correlation, and lower root-mean-square error. The proposed model holds promise for improving our understanding and predicting capabilities of the ENSO phenomenon and can be broadly applicable to other weather and climate prediction scenarios with spatial patterns and teleconnections.
We introduce RedMotion, a transformer model for motion prediction in self-driving vehicles that learns environment representations via redundancy reduction. Our first type of redundancy reduction is induced by an internal transformer decoder and reduces a variable-sized set of local road environment tokens, representing road graphs and agent data, to a fixed-sized global embedding. The second type of redundancy reduction is obtained by self-supervised learning and applies the redundancy reduction principle to embeddings generated from augmented views of road environments. Our experiments reveal that our representation learning approach outperforms PreTraM, Traj-MAE, and GraphDINO in a semi-supervised setting. Moreover, RedMotion achieves competitive results compared to HPTR or MTR++ in the Waymo Motion Prediction Challenge. Our open-source implementation is available at: https://github.com/kit-mrt/future-motion