Despite considerable progress being achieved in point cloud geometry compression, there still remains a challenge in effectively compressing large-scale scenes with sparse surfaces. Another key challenge lies in reducing decoding latency, a crucial requirement in real-world application. In this paper, we propose Pointsoup, an efficient learning-based geometry codec that attains high-performance and extremely low-decoding-latency simultaneously. Inspired by conventional Trisoup codec, a point model-based strategy is devised to characterize local surfaces. Specifically, skin features are embedded from local windows via an attention-based encoder, and dilated windows are introduced as cross-scale priors to infer the distribution of quantized features in parallel. During decoding, features undergo fast refinement, followed by a folding-based point generator that reconstructs point coordinates with fairly fast speed. Experiments show that Pointsoup achieves state-of-the-art performance on multiple benchmarks with significantly lower decoding complexity, i.e., up to 90$\sim$160$\times$ faster than the G-PCCv23 Trisoup decoder on a comparatively low-end platform (e.g., one RTX 2080Ti). Furthermore, it offers variable-rate control with a single neural model (2.9MB), which is attractive for industrial practitioners.
Automatic Speech Recognition (ASR) can play a crucial role in enhancing the accessibility of spoken languages worldwide. In this paper, we build a set of ASR tools for Amharic, a language spoken by more than 50 million people primarily in eastern Africa. Amharic is written in the Ge'ez script, a sequence of graphemes with spacings denoting word boundaries. This makes computational processing of Amharic challenging since the location of spacings can significantly impact the meaning of formed sentences. We find that existing benchmarks for Amharic ASR do not account for these spacings and only measure individual grapheme error rates, leading to significantly inflated measurements of in-the-wild performance. In this paper, we first release corrected transcriptions of existing Amharic ASR test datasets, enabling the community to accurately evaluate progress. Furthermore, we introduce a post-processing approach using a transformer encoder-decoder architecture to organize raw ASR outputs into a grammatically complete and semantically meaningful Amharic sentence. Through experiments on the corrected test dataset, our model enhances the semantic correctness of Amharic speech recognition systems, achieving a Character Error Rate (CER) of 5.5\% and a Word Error Rate (WER) of 23.3\%.
Tokenization is widely used in large language models because it significantly improves performance. However, tokenization imposes several disadvantages, such as performance biases, increased adversarial vulnerability, decreased character-level modeling performance, and increased modeling complexity. To address these disadvantages without sacrificing performance, we propose SpaceByte, a novel byte-level decoder architecture that closes the performance gap between byte-level and subword autoregressive language modeling. SpaceByte consists of a byte-level Transformer model, but with extra larger transformer blocks inserted in the middle of the layers. We find that performance is significantly improved by applying these larger blocks only after certain bytes, such as space characters, which typically denote word boundaries. Our experiments show that for a fixed training and inference compute budget, SpaceByte outperforms other byte-level architectures and roughly matches the performance of tokenized Transformer architectures.
Quantitative and numerical comprehension in language is an important task in many fields like education and finance, but still remains a challenging task for language models. While tool and calculator usage has shown to be helpful to improve mathematical reasoning in large pretrained decoder-only language models, this remains unexplored for smaller language models with encoders. In this paper, we propose Pre-Calc, a simple pre-finetuning objective of learning to use the calculator for both encoder-only and encoder-decoder architectures, formulated as a discriminative and generative task respectively. We pre-train BERT and RoBERTa for discriminative calculator use and Flan-T5 for generative calculator use on the MAWPS, SVAMP, and AsDiv-A datasets, which improves performance on downstream tasks that require numerical understanding. Our code and data are available at https://github.com/calc-cmu/pre-calc.
Dental fluorosis is a chronic disease caused by long-term overconsumption of fluoride, which leads to changes in the appearance of tooth enamel. It is an important basis for early non-invasive diagnosis of endemic fluorosis. However, even dental professionals may not be able to accurately distinguish dental fluorosis and its severity based on tooth images. Currently, there is still a gap in research on applying deep learning to diagnosing dental fluorosis. Therefore, we construct the first open-source dental fluorosis image dataset (DFID), laying the foundation for deep learning research in this field. To advance the diagnosis of dental fluorosis, we propose a pioneering deep learning model called masked latent transformer with the random masking ratio (MLTrMR). MLTrMR introduces a mask latent modeling scheme based on Vision Transformer to enhance contextual learning of dental fluorosis lesion characteristics. Consisting of a latent embedder, encoder, and decoder, MLTrMR employs the latent embedder to extract latent tokens from the original image, whereas the encoder and decoder comprising the latent transformer (LT) block are used to process unmasked tokens and predict masked tokens, respectively. To mitigate the lack of inductive bias in Vision Transformer, which may result in performance degradation, the LT block introduces latent tokens to enhance the learning capacity of latent lesion features. Furthermore, we design an auxiliary loss function to constrain the parameter update direction of the model. MLTrMR achieves 80.19% accuracy, 75.79% F1, and 81.28% quadratic weighted kappa on DFID, making it state-of-the-art (SOTA).
Self-supervised learned (SSL) models such as Wav2vec and HuBERT yield state-of-the-art results on speech-related tasks. Given the effectiveness of such models, it is advantageous to use them in conventional ASR systems. While some approaches suggest incorporating these models as a trainable encoder or a learnable frontend, training such systems is extremely slow and requires a lot of computation cycles. In this work, we propose two simple approaches that use (1) framewise addition and (2) cross-attention mechanisms to efficiently incorporate the representations from the SSL model(s) into the ASR architecture, resulting in models that are comparable in size with standard encoder-decoder conformer systems while also avoiding the usage of SSL models during training. Our approach results in faster training and yields significant performance gains on the Librispeech and Tedlium datasets compared to baselines. We further provide detailed analysis and ablation studies that demonstrate the effectiveness of our approach.
Deep learning models have a risk of utilizing spurious clues to make predictions, such as recognizing actions based on the background scene. This issue can severely degrade the open-set action recognition performance when the testing samples have different scene distributions from the training samples. To mitigate this problem, we propose a novel method, called Scene-debiasing Open-set Action Recognition (SOAR), which features an adversarial scene reconstruction module and an adaptive adversarial scene classification module. The former prevents the decoder from reconstructing the video background given video features, and thus helps reduce the background information in feature learning. The latter aims to confuse scene type classification given video features, with a specific emphasis on the action foreground, and helps to learn scene-invariant information. In addition, we design an experiment to quantify the scene bias. The results indicate that the current open-set action recognizers are biased toward the scene, and our proposed SOAR method better mitigates such bias. Furthermore, our extensive experiments demonstrate that our method outperforms state-of-the-art methods, and the ablation studies confirm the effectiveness of our proposed modules.
Multi-task networks can potentially improve performance and computational efficiency compared to single-task networks, facilitating online deployment. However, current multi-task architectures in point cloud perception combine multiple task-specific point cloud representations, each requiring a separate feature encoder and making the network structures bulky and slow. We propose PAttFormer, an efficient multi-task architecture for joint semantic segmentation and object detection in point clouds that only relies on a point-based representation. The network builds on transformer-based feature encoders using neighborhood attention and grid-pooling and a query-based detection decoder using a novel 3D deformable-attention detection head design. Unlike other LiDAR-based multi-task architectures, our proposed PAttFormer does not require separate feature encoders for multiple task-specific point cloud representations, resulting in a network that is 3x smaller and 1.4x faster while achieving competitive performance on the nuScenes and KITTI benchmarks for autonomous driving perception. Our extensive evaluations show substantial gains from multi-task learning, improving LiDAR semantic segmentation by +1.7% in mIou and 3D object detection by +1.7% in mAP on the nuScenes benchmark compared to the single-task models.
With the rapid development of OCR technology, mixed-scene text recognition has become a key technical challenge. Although deep learning models have achieved significant results in specific scenarios, their generality and stability still need improvement, and the high demand for computing resources affects flexibility. To address these issues, this paper proposes DLoRA-TrOCR, a parameter-efficient hybrid text spotting method based on a pre-trained OCR Transformer. By embedding a weight-decomposed DoRA module in the image encoder and a LoRA module in the text decoder, this method can be efficiently fine-tuned on various downstream tasks. Our method requires no more than 0.7\% trainable parameters, not only accelerating the training efficiency but also significantly improving the recognition accuracy and cross-dataset generalization performance of the OCR system in mixed text scenes. Experiments show that our proposed DLoRA-TrOCR outperforms other parameter-efficient fine-tuning methods in recognizing complex scenes with mixed handwritten, printed, and street text, achieving a CER of 4.02 on the IAM dataset, a F1 score of 94.29 on the SROIE dataset, and a WAR of 86.70 on the STR Benchmark, reaching state-of-the-art performance.
Source camera identification (SCI) technology has attracted increasing attentions over the past few years. However, the existing methods suppress image content with denoising filters that are largely agnostic to the specific sensor pattern noise (SPN) signal of interest. Such practices may potentially degrade the performance of SPN-based SCI due to un-reliable SPNs, especially when forensic images are transmitted through social networking platforms. In this paper, we address the problem of SPN-based device identification and propose a multi-scale feature fusion network (MSFFN) to boost the sensor-based source camera identification attribution. Specifically, several image patches of different scales are selected and input into the MSFFN to extract the SPN. The MSFFN is a multi-scale encoder-decoder structure, which is used to suppress image content and improve source attribution. Subsequently, the content-independent SPN features of different scales are fused. At last, the fused features are used for image source identification. Experimental results compared with the state-of-the-art demonstrate that the proposed scheme achieves significant improvements, especially in the accuracy of social networking image source identification.
Virtual network embedding (VNE) is an essential resource allocation task in network virtualization, aiming to map virtual network requests (VNRs) onto physical infrastructure. Reinforcement learning (RL) has recently emerged as a promising solution to this problem. However, existing RL-based VNE methods are limited by the unidirectional action design and one-size-fits-all training strategy, resulting in restricted searchability and generalizability. In this paper, we propose a FLexible And Generalizable RL framework for VNE, named FlagVNE. Specifically, we design a bidirectional action-based Markov decision process model that enables the joint selection of virtual and physical nodes, thus improving the exploration flexibility of solution space. To tackle the expansive and dynamic action space, we design a hierarchical decoder to generate adaptive action probability distributions and ensure high training efficiency. Furthermore, to overcome the generalization issue for varying VNR sizes, we propose a meta-RL-based training method with a curriculum scheduling strategy, facilitating specialized policy training for each VNR size. Finally, extensive experimental results show the effectiveness of FlagVNE across multiple key metrics. Our code is available at GitHub (https://github.com/GeminiLight/flag-vne).
Vehicular trajectory data from geolocation telematics is vital for analyzing urban mobility patterns. Map-matching aligns noisy, sparsely sampled GPS trajectories with digital road maps to reconstruct accurate vehicle paths. Traditional methods rely on geometric proximity, topology, and shortest-path heuristics, but they overlook two key factors: (1) drivers may prefer routes based on local road characteristics rather than shortest paths, revealing learnable shared preferences, and (2) GPS noise varies spatially due to multipath effects. These factors can reduce the effectiveness of conventional methods in complex scenarios and increase the effort required for heuristic-based implementations. This study introduces a data-driven, deep learning-based map-matching framework, formulating the task as machine translation, inspired by NLP. Specifically, a transformer-based encoder-decoder model learns contextual representations of noisy GPS points to infer trajectory behavior and road structures in an end-to-end manner. Trained on large-scale trajectory data, the method improves path estimation accuracy. Experiments on synthetic trajectories show that this approach outperforms conventional methods by integrating contextual awareness. Evaluation on real-world GPS traces from Manhattan, New York, achieves 75% accuracy in reconstructing navigated routes. These results highlight the effectiveness of transformers in capturing drivers' trajectory behaviors, spatial dependencies, and noise patterns, offering a scalable, robust solution for map-matching. This work contributes to advancing trajectory-driven foundation models for geospatial modeling and urban mobility applications.
Satellite communications can provide massive connections and seamless coverage, but they also face several challenges, such as rain attenuation, long propagation delays, and co-channel interference. To improve transmission efficiency and address severe scenarios, semantic communication has become a popular choice, particularly when equipped with foundation models (FMs). In this study, we introduce an FM-based semantic satellite communication framework, termed FMSAT. This framework leverages FM-based segmentation and reconstruction to significantly reduce bandwidth requirements and accurately recover semantic features under high noise and interference. Considering the high speed of satellites, an adaptive encoder-decoder is proposed to protect important features and avoid frequent retransmissions. Meanwhile, a well-received image can provide a reference for repairing damaged images under sudden attenuation. Since acknowledgment feedback is subject to long propagation delays when retransmission is unavoidable, a novel error detection method is proposed to roughly detect semantic errors at the regenerative satellite. With the proposed detectors at both the satellite and the gateway, the quality of the received images can be ensured. The simulation results demonstrate that the proposed method can significantly reduce bandwidth requirements, adapt to complex satellite scenarios, and protect semantic information with an acceptable transmission delay.
Most advanced unsupervised anomaly detection (UAD) methods rely on modeling feature representations of frozen encoder networks pre-trained on large-scale datasets, e.g. ImageNet. However, the features extracted from the encoders that are borrowed from natural image domains coincide little with the features required in the target UAD domain, such as industrial inspection and medical imaging. In this paper, we propose a novel epistemic UAD method, namely ReContrast, which optimizes the entire network to reduce biases towards the pre-trained image domain and orients the network in the target domain. We start with a feature reconstruction approach that detects anomalies from errors. Essentially, the elements of contrastive learning are elegantly embedded in feature reconstruction to prevent the network from training instability, pattern collapse, and identical shortcut, while simultaneously optimizing both the encoder and decoder on the target domain. To demonstrate our transfer ability on various image domains, we conduct extensive experiments across two popular industrial defect detection benchmarks and three medical image UAD tasks, which shows our superiority over current state-of-the-art methods.
In this paper we investigate the use of decoder-based generative transformers for extracting sentiment towards the named entities in Russian news articles. We study sentiment analysis capabilities of instruction-tuned large language models (LLMs). We consider the dataset of RuSentNE-2023 in our study. The first group of experiments was aimed at the evaluation of zero-shot capabilities of LLMs with closed and open transparencies. The second covers the fine-tuning of Flan-T5 using the "chain-of-thought" (CoT) three-hop reasoning framework (THoR). We found that the results of the zero-shot approaches are similar to the results achieved by baseline fine-tuned encoder-based transformers (BERT-base). Reasoning capabilities of the fine-tuned Flan-T5 models with THoR achieve at least 5% increment with the base-size model compared to the results of the zero-shot experiment. The best results of sentiment analysis on RuSentNE-2023 were achieved by fine-tuned Flan-T5-xl, which surpassed the results of previous state-of-the-art transformer-based classifiers. Our CoT application framework is publicly available: https://github.com/nicolay-r/Reasoning-for-Sentiment-Analysis-Framework
Change detection (CD) from remote sensing (RS) images using deep learning has been widely investigated in the literature. It is typically regarded as a pixel-wise labeling task that aims to classify each pixel as changed or unchanged. Although per-pixel classification networks in encoder-decoder structures have shown dominance, they still suffer from imprecise boundaries and incomplete object delineation at various scenes. For high-resolution RS images, partly or totally changed objects are more worthy of attention rather than a single pixel. Therefore, we revisit the CD task from the mask prediction and classification perspective and propose MaskCD to detect changed areas by adaptively generating categorized masks from input image pairs. Specifically, it utilizes a cross-level change representation perceiver (CLCRP) to learn multiscale change-aware representations and capture spatiotemporal relations from encoded features by exploiting deformable multihead self-attention (DeformMHSA). Subsequently, a masked-attention-based detection transformers (MA-DETR) decoder is developed to accurately locate and identify changed objects based on masked attention and self-attention mechanisms. It reconstructs the desired changed objects by decoding the pixel-wise representations into learnable mask proposals and making final predictions from these candidates. Experimental results on five benchmark datasets demonstrate the proposed approach outperforms other state-of-the-art models. Codes and pretrained models are available online (https://github.com/EricYu97/MaskCD).
With the advancements in 3D printing technologies, it is extremely important that the quality of 3D printed objects, and dimensional accuracies should meet the customer's specifications. Various factors during metal printing affect the printed parts' quality, including the power quality, the printing stage parameters, the print part's location inside the print bed, the curing stage parameters, and the metal sintering process. With the large data gathered from HP's MetJet printing process, AI techniques can be used to analyze, learn, and effectively infer the printed part quality metrics, as well as assist in improving the print yield. In-situ thermal sensing data captured by printer-installed thermal sensors contains the part thermal signature of fusing layers. Such part thermal signature contains a convoluted impact from various factors. In this paper, we use a multimodal thermal encoder network to fuse data of a different nature including the video data vectorized printer control data, and exact part thermal signatures with a trained encoder-decoder module. We explored the data fusing techniques and stages for data fusing, the optimized end-to-end model architecture indicates an improved part quality prediction accuracy.
The emergence of attention-based transformer models has led to their extensive use in various tasks, due to their superior generalization and transfer properties. Recent research has demonstrated that such models, when prompted appropriately, are excellent for few-shot inference. However, such techniques are under-explored for dense prediction tasks like semantic segmentation. In this work, we examine the effectiveness of prompting a transformer-decoder with learned visual prompts for the generalized few-shot segmentation (GFSS) task. Our goal is to achieve strong performance not only on novel categories with limited examples, but also to retain performance on base categories. We propose an approach to learn visual prompts with limited examples. These learned visual prompts are used to prompt a multiscale transformer decoder to facilitate accurate dense predictions. Additionally, we introduce a unidirectional causal attention mechanism between the novel prompts, learned with limited examples, and the base prompts, learned with abundant data. This mechanism enriches the novel prompts without deteriorating the base class performance. Overall, this form of prompting helps us achieve state-of-the-art performance for GFSS on two different benchmark datasets: COCO-$20^i$ and Pascal-$5^i$, without the need for test-time optimization (or transduction). Furthermore, test-time optimization leveraging unlabelled test data can be used to improve the prompts, which we refer to as transductive prompt tuning.
We introduce a novel diffusion transformer, LazyDiffusion, that generates partial image updates efficiently. Our approach targets interactive image editing applications in which, starting from a blank canvas or an image, a user specifies a sequence of localized image modifications using binary masks and text prompts. Our generator operates in two phases. First, a context encoder processes the current canvas and user mask to produce a compact global context tailored to the region to generate. Second, conditioned on this context, a diffusion-based transformer decoder synthesizes the masked pixels in a "lazy" fashion, i.e., it only generates the masked region. This contrasts with previous works that either regenerate the full canvas, wasting time and computation, or confine processing to a tight rectangular crop around the mask, ignoring the global image context altogether. Our decoder's runtime scales with the mask size, which is typically small, while our encoder introduces negligible overhead. We demonstrate that our approach is competitive with state-of-the-art inpainting methods in terms of quality and fidelity while providing a 10x speedup for typical user interactions, where the editing mask represents 10% of the image.
Recent vision foundation models (VFMs) have demonstrated proficiency in various tasks but require supervised fine-tuning to perform the task of semantic segmentation effectively. Benchmarking their performance is essential for selecting current models and guiding future model developments for this task. The lack of a standardized benchmark complicates comparisons. Therefore, the primary objective of this paper is to study how VFMs should be benchmarked for semantic segmentation. To do so, various VFMs are fine-tuned under various settings, and the impact of individual settings on the performance ranking and training time is assessed. Based on the results, the recommendation is to fine-tune the ViT-B variants of VFMs with a 16x16 patch size and a linear decoder, as these settings are representative of using a larger model, more advanced decoder and smaller patch size, while reducing training time by more than 13 times. Using multiple datasets for training and evaluation is also recommended, as the performance ranking across datasets and domain shifts varies. Linear probing, a common practice for some VFMs, is not recommended, as it is not representative of end-to-end fine-tuning. The benchmarking setup recommended in this paper enables a performance analysis of VFMs for semantic segmentation. The findings of such an analysis reveal that pretraining with promptable segmentation is not beneficial, whereas masked image modeling (MIM) with abstract representations is crucial, even more important than the type of supervision used. The code for efficiently fine-tuning VFMs for semantic segmentation can be accessed through the project page at: https://tue-mps.github.io/benchmark-vfm-ss/.