Trajectory prediction for autonomous driving must continuously reason the motion stochasticity of road agents and comply with scene constraints. Existing methods typically rely on one-stage trajectory prediction models, which condition future trajectories on observed trajectories combined with fused scene information. However, they often struggle with complex scene constraints, such as those encountered at intersections. To this end, we present a novel method, called LAformer. It uses a temporally dense lane-aware estimation module to select only the top highly potential lane segments in an HD map, which effectively and continuously aligns motion dynamics with scene information, reducing the representation requirements for the subsequent attention-based decoder by filtering out irrelevant lane segments. Additionally, unlike one-stage prediction models, LAformer utilizes predictions from the first stage as anchor trajectories and adds a second-stage motion refinement module to further explore temporal consistency across the complete time horizon. Extensive experiments on Argoverse 1 and nuScenes demonstrate that LAformer achieves excellent performance for multimodal trajectory prediction.
Knee Osteoarthritis (KOA) is a common musculoskeletal condition that significantly affects mobility and quality of life, particularly in elderly populations. However, training deep learning models for early KOA classification is often hampered by the limited availability of annotated medical datasets, owing to the high costs and labour-intensive nature of data labelling. Traditional data augmentation techniques, while useful, rely on simple transformations and fail to introduce sufficient diversity into the dataset. To address these challenges, we propose the Key-Exchange Convolutional Auto-Encoder (KECAE) as an innovative Artificial Intelligence (AI)-based data augmentation strategy for early KOA classification. Our model employs a convolutional autoencoder with a novel key-exchange mechanism that generates synthetic images by selectively exchanging key pathological features between X-ray images, which not only diversifies the dataset but also ensures the clinical validity of the augmented data. A hybrid loss function is introduced to supervise feature learning and reconstruction, integrating multiple components, including reconstruction, supervision, and feature separation losses. Experimental results demonstrate that the KECAE-generated data significantly improve the performance of KOA classification models, with accuracy gains of up to 1.98% across various standard and state-of-the-art architectures. Furthermore, a clinical validation study involving expert radiologists confirms the anatomical plausibility and diagnostic realism of the synthetic outputs. These findings highlight the potential of KECAE as a robust tool for augmenting medical datasets in early KOA detection.
Past work in natural language processing interpretability focused mainly on popular classification tasks while largely overlooking generation settings, partly due to a lack of dedicated tools. In this work, we introduce Inseq, a Python library to democratize access to interpretability analyses of sequence generation models. Inseq enables intuitive and optimized extraction of models' internal information and feature importance scores for popular decoder-only and encoder-decoder Transformers architectures. We showcase its potential by adopting it to highlight gender biases in machine translation models and locate factual knowledge inside GPT-2. Thanks to its extensible interface supporting cutting-edge techniques such as contrastive feature attribution, Inseq can drive future advances in explainable natural language generation, centralizing good practices and enabling fair and reproducible model evaluations.
Distinguishing among different marine benthic habitat characteristics is of key importance in a wide set of seabed operations ranging from installations of oil rigs to laying networks of cables and monitoring the impact of humans on marine ecosystems. The Side-Scan Sonar (SSS) is a widely used imaging sensor in this regard. It produces high-resolution seafloor maps by logging the intensities of sound waves reflected back from the seafloor. In this work, we leverage these acoustic intensity maps to produce pixel-wise categorization of different seafloor types. We propose a novel architecture adapted from the Vision Transformer (ViT) in an encoder-decoder framework. Further, in doing so, the applicability of ViTs is evaluated on smaller datasets. To overcome the lack of CNN-like inductive biases, thereby making ViTs more conducive to applications in low data regimes, we propose a novel feature extraction module to replace the Multi-layer Perceptron (MLP) block within transformer layers and a novel module to extract multiscale patch embeddings. A lightweight decoder is also proposed to complement this design in order to further boost multiscale feature extraction. With the modified architecture, we achieve state-of-the-art results and also meet real-time computational requirements. We make our code available at ~\url{https://github.com/hayatrajani/s3seg-vit
Previous pitch-controllable text-to-speech (TTS) models rely on directly modeling fundamental frequency, leading to low variance in synthesized speech. To address this issue, we propose PITS, an end-to-end pitch-controllable TTS model that utilizes variational inference to model pitch. Based on VITS, PITS incorporates the Yingram encoder, the Yingram decoder, and adversarial training of pitch-shifted synthesis to achieve pitch-controllability. Experiments demonstrate that PITS generates high-quality speech that is indistinguishable from ground truth speech and has high pitch-controllability without quality degradation. Code, audio samples, and demo are available at https://github.com/anonymous-pits/pits.
Human motion transfer aims to transfer motions from a target dynamic person to a source static one for motion synthesis. An accurate matching between the source person and the target motion in both large and subtle motion changes is vital for improving the transferred motion quality. In this paper, we propose Human MotionFormer, a hierarchical ViT framework that leverages global and local perceptions to capture large and subtle motion matching, respectively. It consists of two ViT encoders to extract input features (i.e., a target motion image and a source human image) and a ViT decoder with several cascaded blocks for feature matching and motion transfer. In each block, we set the target motion feature as Query and the source person as Key and Value, calculating the cross-attention maps to conduct a global feature matching. Further, we introduce a convolutional layer to improve the local perception after the global cross-attention computations. This matching process is implemented in both warping and generation branches to guide the motion transfer. During training, we propose a mutual learning loss to enable the co-supervision between warping and generation branches for better motion representations. Experiments show that our Human MotionFormer sets the new state-of-the-art performance both qualitatively and quantitatively. Project page: \url{https://github.com/KumapowerLIU/Human-MotionFormer}
This paper presents a deep learning framework for medical video segmentation. Convolution neural network (CNN) and transformer-based methods have achieved great milestones in medical image segmentation tasks due to their incredible semantic feature encoding and global information comprehension abilities. However, most existing approaches ignore a salient aspect of medical video data - the temporal dimension. Our proposed framework explicitly extracts features from neighbouring frames across the temporal dimension and incorporates them with a temporal feature blender, which then tokenises the high-level spatio-temporal feature to form a strong global feature encoded via a Swin Transformer. The final segmentation results are produced via a UNet-like encoder-decoder architecture. Our model outperforms other approaches by a significant margin and improves the segmentation benchmarks on the VFSS2022 dataset, achieving a dice coefficient of 0.8986 and 0.8186 for the two datasets tested. Our studies also show the efficacy of the temporal feature blending scheme and cross-dataset transferability of learned capabilities. Code and models are fully available at https://github.com/SimonZeng7108/Video-SwinUNet.
In this paper, we propose an iterative source error correction (ISEC) decoding scheme for deep-learning-based joint source-channel coding (Deep JSCC). Given a noisy codeword received through the channel, we use a Deep JSCC encoder and decoder pair to update the codeword iteratively to find a (modified) maximum a-posteriori (MAP) solution. For efficient MAP decoding, we utilize a neural network-based denoiser to approximate the gradient of the log-prior density of the codeword space. Albeit the non-convexity of the optimization problem, our proposed scheme improves various distortion and perceptual quality metrics from the conventional one-shot (non-iterative) Deep JSCC decoding baseline. Furthermore, the proposed scheme produces more reliable source reconstruction results compared to the baseline when the channel noise characteristics do not match the ones used during training.
Image restoration is a long-standing low-level vision problem, e.g., deblurring and deraining. In the process of image restoration, it is necessary to consider not only the spatial details and contextual information of restoration to ensure the quality, but also the system complexity. Although many methods have been able to guarantee the quality of image restoration, the system complexity of the state-of-the-art (SOTA) methods is increasing as well. Motivated by this, we present a mixed hierarchy network that can balance these competing goals. Our main proposal is a mixed hierarchy architecture, that progressively recovers contextual information and spatial details from degraded images while we design intra-blocks to reduce system complexity. Specifically, our model first learns the contextual information using encoder-decoder architectures, and then combines them with high-resolution branches that preserve spatial detail. In order to reduce the system complexity of this architecture for convenient analysis and comparison, we replace or remove the nonlinear activation function with multiplication and use a simple network structure. In addition, we replace spatial convolution with global self-attention for the middle block of encoder-decoder. The resulting tightly interlinked hierarchy architecture, named as MHNet, delivers strong performance gains on several image restoration tasks, including image deraining, and deblurring.
Large pre-trained language models (PLMs) have been shown to retain implicit knowledge within their parameters. To enhance this implicit knowledge, we propose Knowledge Injection into Language Models (KILM), a novel approach that injects entity-related knowledge into encoder-decoder PLMs, via a generative knowledge infilling objective through continued pre-training. This is done without architectural modifications to the PLMs or adding additional parameters. Experimental results over a suite of knowledge-intensive tasks spanning numerous datasets show that KILM enables models to retain more knowledge and hallucinate less, while preserving their original performance on general NLU and NLG tasks. KILM also demonstrates improved zero-shot performances on tasks such as entity disambiguation, outperforming state-of-the-art models having 30x more parameters.
Inspired by retrieval-augmented language generation and pretrained Vision and Language (V&L) encoders, we present a new approach to image captioning that generates sentences given the input image and a set of captions retrieved from a datastore, as opposed to the image alone. The encoder in our model jointly processes the image and retrieved captions using a pretrained V&L BERT, while the decoder attends to the multimodal encoder representations, benefiting from the extra textual evidence from the retrieved captions. Experimental results on the COCO dataset show that image captioning can be effectively formulated from this new perspective. Our model, named EXTRA, benefits from using captions retrieved from the training dataset, and it can also benefit from using an external dataset without the need for retraining. Ablation studies show that retrieving a sufficient number of captions (e.g., k=5) can improve captioning quality. Our work contributes towards using pretrained V&L encoders for generative tasks, instead of standard classification tasks.
In this work, instead of directly predicting the pixel-level segmentation masks, the problem of referring image segmentation is formulated as sequential polygon generation, and the predicted polygons can be later converted into segmentation masks. This is enabled by a new sequence-to-sequence framework, Polygon Transformer (PolyFormer), which takes a sequence of image patches and text query tokens as input, and outputs a sequence of polygon vertices autoregressively. For more accurate geometric localization, we propose a regression-based decoder, which predicts the precise floating-point coordinates directly, without any coordinate quantization error. In the experiments, PolyFormer outperforms the prior art by a clear margin, e.g., 5.40% and 4.52% absolute improvements on the challenging RefCOCO+ and RefCOCOg datasets. It also shows strong generalization ability when evaluated on the referring video segmentation task without fine-tuning, e.g., achieving competitive 61.5% J&F on the Ref-DAVIS17 dataset.
Significant developments in techniques such as encoder-decoder models have enabled us to represent information comprising multiple modalities. This information can further enhance many downstream tasks in the field of information retrieval and natural language processing; however, improvements in multi-modal techniques and their performance evaluation require large-scale multi-modal data which offers sufficient diversity. Multi-lingual modeling for a variety of tasks like multi-modal summarization, text generation, and translation leverages information derived from high-quality multi-lingual annotated data. In this work, we present the current largest multi-lingual multi-modal summarization dataset (M3LS), and it consists of over a million instances of document-image pairs along with a professionally annotated multi-modal summary for each pair. It is derived from news articles published by British Broadcasting Corporation(BBC) over a decade and spans 20 languages, targeting diversity across five language roots, it is also the largest summarization dataset for 13 languages and consists of cross-lingual summarization data for 2 languages. We formally define the multi-lingual multi-modal summarization task utilizing our dataset and report baseline scores from various state-of-the-art summarization techniques in a multi-lingual setting. We also compare it with many similar datasets to analyze the uniqueness and difficulty of M3LS.
One of the recent best attempts at Text-to-SQL is the pre-trained language model. Due to the structural property of the SQL queries, the seq2seq model takes the responsibility of parsing both the schema items (i.e., tables and columns) and the skeleton (i.e., SQL keywords). Such coupled targets increase the difficulty of parsing the correct SQL queries especially when they involve many schema items and logic operators. This paper proposes a ranking-enhanced encoding and skeleton-aware decoding framework to decouple the schema linking and the skeleton parsing. Specifically, for a seq2seq encoder-decode model, its encoder is injected by the most relevant schema items instead of the whole unordered ones, which could alleviate the schema linking effort during SQL parsing, and its decoder first generates the skeleton and then the actual SQL query, which could implicitly constrain the SQL parsing. We evaluate our proposed framework on Spider and its three robustness variants: Spider-DK, Spider-Syn, and Spider-Realistic. The experimental results show that our framework delivers promising performance and robustness. Our code is available at https://github.com/RUCKBReasoning/RESDSQL.
Multi-scale features are essential for dense prediction tasks, such as object detection, instance segmentation, and semantic segmentation. The prevailing methods usually utilize a classification backbone to extract multi-scale features and then fuse these features using a lightweight module (e.g., the fusion module in FPN and BiFPN, two typical object detection methods). However, as these methods allocate most computational resources to the classification backbone, the multi-scale feature fusion in these methods is delayed, which may lead to inadequate feature fusion. While some methods perform feature fusion from early stages, they either fail to fully leverage high-level features to guide low-level feature learning or have complex structures, resulting in sub-optimal performance. We propose a streamlined cascade encoder-decoder network, dubbed CEDNet, tailored for dense \mbox{prediction} tasks. All stages in CEDNet share the same encoder-decoder structure and perform multi-scale feature fusion within the decoder. A hallmark of CEDNet is its ability to incorporate high-level features from early stages to guide low-level feature learning in subsequent stages, thereby enhancing the effectiveness of multi-scale feature fusion. We explored three well-known encoder-decoder structures: Hourglass, UNet, and FPN. When integrated into CEDNet, they performed much better than traditional methods that use a pre-designed classification backbone combined with a lightweight fusion module. Extensive experiments on object detection, instance segmentation, and semantic segmentation demonstrated the effectiveness of our method. The code is available at https://github.com/zhanggang001/CEDNet.
Fully-supervised shadow removal methods achieve the best restoration qualities on public datasets but still generate some shadow remnants. One of the reasons is the lack of large-scale shadow & shadow-free image pairs. Unsupervised methods can alleviate the issue but their restoration qualities are much lower than those of fully-supervised methods. In this work, we find that pretraining shadow removal networks on the image inpainting dataset can reduce the shadow remnants significantly: a naive encoder-decoder network gets competitive restoration quality w.r.t. the state-of-the-art methods via only 10% shadow & shadow-free image pairs. After analyzing networks with/without inpainting pre-training via the information stored in the weight (IIW), we find that inpainting pretraining improves restoration quality in non-shadow regions and enhances the generalization ability of networks significantly. Additionally, shadow removal fine-tuning enables networks to fill in the details of shadow regions. Inspired by these observations we formulate shadow removal as an adaptive fusion task that takes advantage of both shadow removal and image inpainting. Specifically, we develop an adaptive fusion network consisting of two encoders, an adaptive fusion block, and a decoder. The two encoders are responsible for extracting the feature from the shadow image and the shadow-masked image respectively. The adaptive fusion block is responsible for combining these features in an adaptive manner. Finally, the decoder converts the adaptive fused features to the desired shadow-free result. The extensive experiments show that our method empowered with inpainting outperforms all state-of-the-art methods.
Neural image compression has surpassed state-of-the-art traditional codecs (H.266/VVC) for rate-distortion (RD) performance, but suffers from large complexity and separate models for different rate-distortion trade-offs. In this paper, we propose an Efficient single-model Variable-bit-rate Codec (EVC), which is able to run at 30 FPS with 768x512 input images and still outperforms VVC for the RD performance. By further reducing both encoder and decoder complexities, our small model even achieves 30 FPS with 1920x1080 input images. To bridge the performance gap between our different capacities models, we meticulously design the mask decay, which transforms the large model's parameters into the small model automatically. And a novel sparsity regularization loss is proposed to mitigate shortcomings of $L_p$ regularization. Our algorithm significantly narrows the performance gap by 50% and 30% for our medium and small models, respectively. At last, we advocate the scalable encoder for neural image compression. The encoding complexity is dynamic to meet different latency requirements. We propose decaying the large encoder multiple times to reduce the residual representation progressively. Both mask decay and residual representation learning greatly improve the RD performance of our scalable encoder. Our code is at https://github.com/microsoft/DCVC.
Multilingual machine translation (MMT) benefits from cross-lingual transfer but is a challenging multitask optimization problem. This is partly because there is no clear framework to systematically learn language-specific parameters. Self-supervised learning (SSL) approaches that leverage large quantities of monolingual data (where parallel data is unavailable) have shown promise by improving translation performance as complementary tasks to the MMT task. However, jointly optimizing SSL and MMT tasks is even more challenging. In this work, we first investigate how to utilize intra-distillation to learn more *language-specific* parameters and then show the importance of these language-specific parameters. Next, we propose a novel but simple SSL task, concurrent denoising, that co-trains with the MMT task by concurrently denoising monolingual data on both the encoder and decoder. Finally, we apply intra-distillation to this co-training approach. Combining these two approaches significantly improves MMT performance, outperforming three state-of-the-art SSL methods by a large margin, e.g., 11.3\% and 3.7\% improvement on an 8-language and a 15-language benchmark compared with MASS, respectively
Current AI-assisted protein design mainly utilizes protein sequential and structural information. Meanwhile, there exists tremendous knowledge curated by humans in the text format describing proteins' high-level functionalities. Yet, whether the incorporation of such text data can help protein design tasks has not been explored. To bridge this gap, we propose ProteinDT, a multi-modal framework that leverages textual descriptions for protein design. ProteinDT consists of three subsequent steps: ProteinCLAP which aligns the representation of two modalities, a facilitator that generates the protein representation from the text modality, and a decoder that creates the protein sequences from the representation. To train ProteinDT, we construct a large dataset, SwissProtCLAP, with 441K text and protein pairs. We quantitatively verify the effectiveness of ProteinDT on three challenging tasks: (1) over 90% accuracy for text-guided protein generation; (2) best hit ratio on 12 zero-shot text-guided protein editing tasks; (3) superior performance on four out of six protein property prediction benchmarks.
In this paper, we propose an encoder-decoder neural architecture (called Channelformer) to achieve improved channel estimation for orthogonal frequency-division multiplexing (OFDM) waveforms in downlink scenarios. The self-attention mechanism is employed to achieve input precoding for the input features before processing them in the decoder. In particular, we implement multi-head attention in the encoder and a residual convolutional neural architecture as the decoder, respectively. We also employ a customized weight-level pruning to slim the trained neural network with a fine-tuning process, which reduces the computational complexity significantly to realize a low complexity and low latency solution. This enables reductions of up to 70\% in the parameters, while maintaining an almost identical performance compared with the complete Channelformer. We also propose an effective online training method based on the fifth generation (5G) new radio (NR) configuration for the modern communication systems, which only needs the available information at the receiver for online training. Using industrial standard channel models, the simulations of attention-based solutions show superior estimation performance compared with other candidate neural network methods for channel estimation.