Recent sparse detectors with multiple, e.g. six, decoder layers achieve promising performance but much inference time due to complex heads. Previous works have explored using dense priors as initialization and built one-decoder-layer detectors. Although they gain remarkable acceleration, their performance still lags behind their six-decoder-layer counterparts by a large margin. In this work, we aim to bridge this performance gap while retaining fast speed. We find that the architecture discrepancy between dense and sparse detectors leads to feature conflict, hampering the performance of one-decoder-layer detectors. Thus we propose Adaptive Sparse Anchor Generator (ASAG) which predicts dynamic anchors on patches rather than grids in a sparse way so that it alleviates the feature conflict problem. For each image, ASAG dynamically selects which feature maps and which locations to predict, forming a fully adaptive way to generate image-specific anchors. Further, a simple and effective Query Weighting method eases the training instability from adaptiveness. Extensive experiments show that our method outperforms dense-initialized ones and achieves a better speed-accuracy trade-off. The code is available at \url{https://github.com/iSEE-Laboratory/ASAG}.
Nowadays, transformer networks have demonstrated superior performance in many computer vision tasks. In a multi-view 3D reconstruction algorithm following this paradigm, self-attention processing has to deal with intricate image tokens including massive information when facing heavy amounts of view input. The curse of information content leads to the extreme difficulty of model learning. To alleviate this problem, recent methods compress the token number representing each view or discard the attention operations between the tokens from different views. Obviously, they give a negative impact on performance. Therefore, we propose long-range grouping attention (LGA) based on the divide-and-conquer principle. Tokens from all views are grouped for separate attention operations. The tokens in each group are sampled from all views and can provide macro representation for the resided view. The richness of feature learning is guaranteed by the diversity among different groups. An effective and efficient encoder can be established which connects inter-view features using LGA and extract intra-view features using the standard self-attention layer. Moreover, a novel progressive upsampling decoder is also designed for voxel generation with relatively high resolution. Hinging on the above, we construct a powerful transformer-based network, called LRGT. Experimental results on ShapeNet verify our method achieves SOTA accuracy in multi-view reconstruction. Code will be available at https://github.com/LiyingCV/Long-Range-Grouping-Transformer.
Query-based object detectors directly decode image features into object instances with a set of learnable queries. These query vectors are progressively refined to stable meaningful representations through a sequence of decoder layers, and then used to directly predict object locations and categories with simple FFN heads. In this paper, we present a new query-based object detector (DEQDet) by designing a deep equilibrium decoder. Our DEQ decoder models the query vector refinement as the fixed point solving of an {implicit} layer and is equivalent to applying {infinite} steps of refinement. To be more specific to object decoding, we use a two-step unrolled equilibrium equation to explicitly capture the query vector refinement. Accordingly, we are able to incorporate refinement awareness into the DEQ training with the inexact gradient back-propagation (RAG). In addition, to stabilize the training of our DEQDet and improve its generalization ability, we devise the deep supervision scheme on the optimization path of DEQ with refinement-aware perturbation~(RAP). Our experiments demonstrate DEQDet converges faster, consumes less memory, and achieves better results than the baseline counterpart (AdaMixer). In particular, our DEQDet with ResNet50 backbone and 300 queries achieves the $49.5$ mAP and $33.0$ AP$_s$ on the MS COCO benchmark under $2\times$ training scheme (24 epochs).
The Bavarian Academy of Sciences and Humanities aims to digitize its Medieval Latin Dictionary. This dictionary entails record cards referring to lemmas in medieval Latin, a low-resource language. A crucial step of the digitization process is the Handwritten Text Recognition (HTR) of the handwritten lemmas found on these record cards. In our work, we introduce an end-to-end pipeline, tailored to the medieval Latin dictionary, for locating, extracting, and transcribing the lemmas. We employ two state-of-the-art (SOTA) image segmentation models to prepare the initial data set for the HTR task. Furthermore, we experiment with different transformer-based models and conduct a set of experiments to explore the capabilities of different combinations of vision encoders with a GPT-2 decoder. Additionally, we also apply extensive data augmentation resulting in a highly competitive model. The best-performing setup achieved a Character Error Rate (CER) of 0.015, which is even superior to the commercial Google Cloud Vision model, and shows more stable performance.
Compressed prompts aid instruction-tuned language models (LMs) in overcoming context window limitations and reducing computational costs. Existing methods, which primarily based on training embeddings, face various challenges associated with interpretability, the fixed number of embedding tokens, reusability across different LMs, and inapplicability when interacting with black-box APIs. This study proposes prompt compression with reinforcement learning (PCRL), which is a discrete prompt compression method that addresses these issues. The proposed PCRL method utilizes a computationally efficient policy network that edits prompts directly. The training approach employed in the proposed PCRLs can be applied flexibly to various types of LMs, including both decoder-only and encoder-decoder architecture and it can be trained without gradient access to the LMs or labeled data. The proposed PCRL achieves an average reduction of 24.6% in terms of the token count across various instruction prompts while maintaining sufficient performance. In addition, we demonstrate that the learned policy can be transferred to larger LMs, and through a comprehensive analysis, we explore the token importance within the prompts. Our code is accessible at https://github.com/nenomigami/PromptCompressor.
Recent years have seen a growing interest in Scene Graph Generation (SGG), a comprehensive visual scene understanding task that aims to predict entity relationships using a relation encoder-decoder pipeline stacked on top of an object encoder-decoder backbone. Unfortunately, current SGG methods suffer from an information loss regarding the entities local-level cues during the relation encoding process. To mitigate this, we introduce the Vision rElation TransfOrmer (VETO), consisting of a novel local-level entity relation encoder. We further observe that many existing SGG methods claim to be unbiased, but are still biased towards either head or tail classes. To overcome this bias, we introduce a Mutually Exclusive ExperT (MEET) learning strategy that captures important relation features without bias towards head or tail classes. Experimental results on the VG and GQA datasets demonstrate that VETO + MEET boosts the predictive performance by up to 47 percentage over the state of the art while being 10 times smaller.
Polyps segmentation poses a significant challenge in medical imaging due to the flat surface of polyps and their texture similarity to surrounding tissues. This similarity gives rise to difficulties in establishing a clear boundary between polyps and the surrounding mucosa, leading to complications such as local overexposure and the presence of bright spot reflections in imaging. To counter this problem, we propose a new dual graph convolution network (Dual-GCN) and location self-attention mechanisms with weighted fast normalization fusion model, named DSFNet. First, we introduce a feature enhancement block module based on Dual-GCN module to enhance local spatial and structural information extraction with fine granularity. Second, we introduce a location fused self-attention module to enhance the model's awareness and capacity to capture global information. Finally, the weighted fast normalized fusion method with trainable weights is introduced to efficiently integrate the feature maps from encoder, bottleneck, and decoder, thus promoting information transmission and facilitating the semantic consistency. Experimental results show that the proposed model surpasses other state-of-the-art models in gold standard indicators, such as Dice, MAE, and IoU. Both quantitative and qualitative analysis indicate that the proposed model demonstrates exceptional capability in polyps segmentation and has great potential clinical significance. We have shared our code on anonymous website for evaluation.
Connectionist temporal classification (CTC) and attention-based encoder decoder (AED) joint training has been widely applied in automatic speech recognition (ASR). Unlike most hybrid models that separately calculate the CTC and AED losses, our proposed integrated-CTC utilizes the attention mechanism of AED to guide the output of CTC. In this paper, we employ two fusion methods, namely direct addition of logits (DAL) and preserving the maximum probability (PMP). We achieve dimensional consistency by adaptively affine transforming the attention results to match the dimensions of CTC. To accelerate model convergence and improve accuracy, we introduce auxiliary loss regularization for accelerated convergence. Experimental results demonstrate that the DAL method performs better in attention rescoring, while the PMP method excels in CTC prefix beam search and greedy search.
Lake extraction from remote sensing imagery is a complex challenge due to the varied lake shapes and data noise. Current methods rely on multispectral image datasets, making it challenging to learn lake features accurately from pixel arrangements. This, in turn, affects model learning and the creation of accurate segmentation masks. This paper introduces a prompt-based dataset construction approach that provides approximate lake locations using point, box, and mask prompts. We also propose a two-stage prompt enhancement framework, LEPrompter, with prompt-based and prompt-free stages during training. The prompt-based stage employs a prompt encoder to extract prior information, integrating prompt tokens and image embedding through self- and cross-attention in the prompt decoder. Prompts are deactivated to ensure independence during inference, enabling automated lake extraction without introducing additional parameters and GFlops. Extensive experiments showcase performance improvements of our proposed approach compared to the previous state-of-the-art method. The source code is available at https://github.com/BastianChen/LEPrompter.
Video-language pre-trained models have shown remarkable success in guiding video question-answering (VideoQA) tasks. However, due to the length of video sequences, training large-scale video-based models incurs considerably higher costs than training image-based ones. This motivates us to leverage the knowledge from image-based pretraining, despite the obvious gaps between image and video domains. To bridge these gaps, in this paper, we propose Tem-Adapter, which enables the learning of temporal dynamics and complex semantics by a visual Temporal Aligner and a textual Semantic Aligner. Unlike conventional pretrained knowledge adaptation methods that only concentrate on the downstream task objective, the Temporal Aligner introduces an extra language-guided autoregressive task aimed at facilitating the learning of temporal dependencies, with the objective of predicting future states based on historical clues and language guidance that describes event progression. Besides, to reduce the semantic gap and adapt the textual representation for better event description, we introduce a Semantic Aligner that first designs a template to fuse question and answer pairs as event descriptions and then learns a Transformer decoder with the whole video sequence as guidance for refinement. We evaluate Tem-Adapter and different pre-train transferring methods on two VideoQA benchmarks, and the significant performance improvement demonstrates the effectiveness of our method.
Ultrasound (US) imaging is indispensable in clinical practice. To diagnose certain diseases, sonographers must observe corresponding dynamic anatomic structures to gather comprehensive information. However, the limited availability of specific US video cases causes teaching difficulties in identifying corresponding diseases, which potentially impacts the detection rate of such cases. The synthesis of US videos may represent a promising solution to this issue. Nevertheless, it is challenging to accurately animate the intricate motion of dynamic anatomic structures while preserving image fidelity. To address this, we present a novel online feature-decoupling framework called OnUVS for high-fidelity US video synthesis. Our highlights can be summarized by four aspects. First, we introduced anatomic information into keypoint learning through a weakly-supervised training strategy, resulting in improved preservation of anatomical integrity and motion while minimizing the labeling burden. Second, to better preserve the integrity and textural information of US images, we implemented a dual-decoder that decouples the content and textural features in the generator. Third, we adopted a multiple-feature discriminator to extract a comprehensive range of visual cues, thereby enhancing the sharpness and fine details of the generated videos. Fourth, we constrained the motion trajectories of keypoints during online learning to enhance the fluidity of generated videos. Our validation and user studies on in-house echocardiographic and pelvic floor US videos showed that OnUVS synthesizes US videos with high fidelity.
Long-tailed distribution of semantic categories, which has been often ignored in conventional methods, causes unsatisfactory performance in semantic segmentation on tail categories. In this paper, we focus on the problem of long-tailed semantic segmentation. Although some long-tailed recognition methods (e.g., re-sampling/re-weighting) have been proposed in other problems, they can probably compromise crucial contextual information and are thus hardly adaptable to the problem of long-tailed semantic segmentation. To address this issue, we propose MEDOE, a novel framework for long-tailed semantic segmentation via contextual information ensemble-and-grouping. The proposed two-sage framework comprises a multi-expert decoder (MED) and a multi-expert output ensemble (MOE). Specifically, the MED includes several "experts". Based on the pixel frequency distribution, each expert takes the dataset masked according to the specific categories as input and generates contextual information self-adaptively for classification; The MOE adopts learnable decision weights for the ensemble of the experts' outputs. As a model-agnostic framework, our MEDOE can be flexibly and efficiently coupled with various popular deep neural networks (e.g., DeepLabv3+, OCRNet, and PSPNet) to improve their performance in long-tailed semantic segmentation. Experimental results show that the proposed framework outperforms the current methods on both Cityscapes and ADE20K datasets by up to 1.78% in mIoU and 5.89% in mAcc.
We propose an agglomerative Transformer (AGER) that enables Transformer-based human-object interaction (HOI) detectors to flexibly exploit extra instance-level cues in a single-stage and end-to-end manner for the first time. AGER acquires instance tokens by dynamically clustering patch tokens and aligning cluster centers to instances with textual guidance, thus enjoying two benefits: 1) Integrality: each instance token is encouraged to contain all discriminative feature regions of an instance, which demonstrates a significant improvement in the extraction of different instance-level cues and subsequently leads to a new state-of-the-art performance of HOI detection with 36.75 mAP on HICO-Det. 2) Efficiency: the dynamical clustering mechanism allows AGER to generate instance tokens jointly with the feature learning of the Transformer encoder, eliminating the need of an additional object detector or instance decoder in prior methods, thus allowing the extraction of desirable extra cues for HOI detection in a single-stage and end-to-end pipeline. Concretely, AGER reduces GFLOPs by 8.5% and improves FPS by 36%, even compared to a vanilla DETR-like pipeline without extra cue extraction.
We introduce an object-aware decoder for improving the performance of spatio-temporal representations on ego-centric videos. The key idea is to enhance object-awareness during training by tasking the model to predict hand positions, object positions, and the semantic label of the objects using paired captions when available. At inference time the model only requires RGB frames as inputs, and is able to track and ground objects (although it has not been trained explicitly for this). We demonstrate the performance of the object-aware representations learnt by our model, by: (i) evaluating it for strong transfer, i.e. through zero-shot testing, on a number of downstream video-text retrieval and classification benchmarks; and (ii) by using the representations learned as input for long-term video understanding tasks (e.g. Episodic Memory in Ego4D). In all cases the performance improves over the state of the art -- even compared to networks trained with far larger batch sizes. We also show that by using noisy image-level detection as pseudo-labels in training, the model learns to provide better bounding boxes using video consistency, as well as grounding the words in the associated text descriptions. Overall, we show that the model can act as a drop-in replacement for an ego-centric video model to improve performance through visual-text grounding.
The latest advancements in neural image compression show great potential in surpassing the rate-distortion performance of conventional standard codecs. Nevertheless, there exists an indelible domain gap between the datasets utilized for training (i.e., natural images) and those utilized for inference (e.g., artistic images). Our proposal involves a low-rank adaptation approach aimed at addressing the rate-distortion drop observed in out-of-domain datasets. Specifically, we perform low-rank matrix decomposition to update certain adaptation parameters of the client's decoder. These updated parameters, along with image latents, are encoded into a bitstream and transmitted to the decoder in practical scenarios. Due to the low-rank constraint imposed on the adaptation parameters, the resulting bit rate overhead is small. Furthermore, the bit rate allocation of low-rank adaptation is \emph{non-trivial}, considering the diverse inputs require varying adaptation bitstreams. We thus introduce a dynamic gating network on top of the low-rank adaptation method, in order to decide which decoder layer should employ adaptation. The dynamic adaptation network is optimized end-to-end using rate-distortion loss. Our proposed method exhibits universality across diverse image datasets. Extensive results demonstrate that this paradigm significantly mitigates the domain gap, surpassing non-adaptive methods with an average BD-rate improvement of approximately $19\%$ across out-of-domain images. Furthermore, it outperforms the most advanced instance adaptive methods by roughly $5\%$ BD-rate. Ablation studies confirm our method's ability to universally enhance various image compression architectures.
Recent advancements in large foundation models have shown promising potential in the medical industry due to their flexible prompting capability. One such model, the Segment Anything Model (SAM), a prompt-driven segmentation model, has shown remarkable performance improvements, surpassing state-of-the-art approaches in medical image segmentation. However, existing methods primarily rely on tuning strategies that require extensive data or prior prompts tailored to the specific task, making it particularly challenging when only a limited number of data samples are available. In this paper, we propose a novel perspective on self-prompting in medical vision applications. Specifically, we harness the embedding space of SAM to prompt itself through a simple yet effective linear pixel-wise classifier. By preserving the encoding capabilities of the large model, the contextual information from its decoder, and leveraging its interactive promptability, we achieve competitive results on multiple datasets (i.e. improvement of more than 15% compared to fine-tuning the mask decoder using a few images).
The generation of synthetic medical records using Generative Adversarial Networks (GANs) is becoming crucial for addressing privacy concerns and facilitating data sharing in the medical domain. In this paper, we introduce a novel method to create synthetic hybrid medical records that combine both image and non-image data, utilizing an auto-encoding GAN (alphaGAN) and a conditional tabular GAN (CTGAN). Our methodology encompasses three primary steps: I) Dimensional reduction of images in a private dataset (pDS) using the pretrained encoder of the {\alpha}GAN, followed by integration with the remaining non-image clinical data to form tabular representations; II) Training the CTGAN on the encoded pDS to produce a synthetic dataset (sDS) which amalgamates encoded image features with non-image clinical data; and III) Reconstructing synthetic images from the image features using the alphaGAN's pretrained decoder. We successfully generated synthetic records incorporating both Chest X-Rays (CXRs) and thirteen non-image clinical variables (comprising seven categorical and six numeric variables). To evaluate the efficacy of the sDS, we designed classification and regression tasks and compared the performance of models trained on pDS and sDS against the pDS test set. Remarkably, by leveraging five times the volume of sDS for training, we achieved classification and regression results that were comparable, if slightly inferior, to those obtained using the native pDS. Our method holds promise for publicly releasing synthetic datasets without undermining the potential for secondary data usage.
Due to the high similarity between camouflaged instances and the background, the recently proposed camouflaged instance segmentation (CIS) faces challenges in accurate localization and instance segmentation. To this end, inspired by query-based transformers, we propose a unified query-based multi-task learning framework for camouflaged instance segmentation, termed UQFormer, which builds a set of mask queries and a set of boundary queries to learn a shared composed query representation and efficiently integrates global camouflaged object region and boundary cues, for simultaneous instance segmentation and instance boundary detection in camouflaged scenarios. Specifically, we design a composed query learning paradigm that learns a shared representation to capture object region and boundary features by the cross-attention interaction of mask queries and boundary queries in the designed multi-scale unified learning transformer decoder. Then, we present a transformer-based multi-task learning framework for simultaneous camouflaged instance segmentation and camouflaged instance boundary detection based on the learned composed query representation, which also forces the model to learn a strong instance-level query representation. Notably, our model views the instance segmentation as a query-based direct set prediction problem, without other post-processing such as non-maximal suppression. Compared with 14 state-of-the-art approaches, our UQFormer significantly improves the performance of camouflaged instance segmentation. Our code will be available at https://github.com/dongbo811/UQFormer.
The problem of packing smaller objects within a larger object has been of interest since decades. In these problems, in addition to the requirement that the smaller objects must lie completely inside the larger objects, they are expected to not overlap or have minimum overlap with each other. Due to this, the problem of packing turns out to be a non-convex problem, obtaining whose optimal solution is challenging. As such, several heuristic approaches have been used for obtaining sub-optimal solutions in general, and provably optimal solutions for some special instances. In this paper, we propose a novel encoder-decoder architecture consisting of an encoder block, a perturbation block and a decoder block, for packing identical circles within a larger circle. In our approach, the encoder takes the index of a circle to be packed as an input and outputs its center through a normalization layer, the perturbation layer adds controlled perturbations to the center, ensuring that it does not deviate beyond the radius of the smaller circle to be packed, and the decoder takes the perturbed center as input and estimates the index of the intended circle for packing. We parameterize the encoder and decoder by a neural network and optimize it to reduce an error between the decoder's estimated index and the actual index of the circle provided as input to the encoder. The proposed approach can be generalized to pack objects of higher dimensions and different shapes by carefully choosing normalization and perturbation layers. The approach gives a sub-optimal solution and is able to pack smaller objects within a larger object with competitive performance with respect to classical methods.
In this paper, we investigate the in-context learning ability of retrieval-augmented encoder-decoder language models. We first conduct a comprehensive analysis of existing models and identify their limitations in in-context learning, primarily due to a mismatch between pretraining and inference, as well as a restricted context length. To address these issues, we propose RAVEN, a model that combines retrieval-augmented masked language modeling and prefix language modeling. We further introduce Fusion-in-Context Learning to enhance the few-shot performance by enabling the model to leverage more in-context examples without requiring additional training. Through extensive experiments, we demonstrate that our simple yet effective design significantly improves performance, achieving results comparable to the most advanced language models in certain scenarios, despite having substantially fewer parameters. Our work underscores the potential of retrieval-augmented encoder-decoder language models for in-context learning and encourages further research in this direction.