Raghuveer Rao

Zhikang Zhang, Zifan Yu, Suya You et al. · amazon-science

Motivated by the increasing application of low-resolution LiDAR recently, we target the problem of low-resolution LiDAR-camera calibration in this work. The main challenges are two-fold: sparsity and noise in point clouds. To address the problem, we propose to apply depth interpolation to increase the point density and supervised contrastive learning to learn noise-resistant features. The experiments on RELLIS-3D demonstrate that our approach achieves an average mean absolute rotation/translation errors of 0.15cm/0.33\textdegree on 32-channel LiDAR point cloud data, which significantly outperforms all reference methods.

Cheng Han, Qifan Wang, Sohail A. Dianat et al.

Transformer-based architectures have become the de-facto standard models for diverse vision tasks owing to their superior performance. As the size of the models continues to scale up, model distillation becomes extremely important in various real applications, particularly on devices limited by computational resources. However, prevailing knowledge distillation methods exhibit diminished efficacy when confronted with a large capacity gap between the teacher and the student, e.g, 10x compression rate. In this paper, we present a novel approach named Automatic Multi-step Distillation (AMD) for large-scale vision model compression. In particular, our distillation process unfolds across multiple steps. Initially, the teacher undergoes distillation to form an intermediate teacher-assistant model, which is subsequently distilled further to the student. An efficient and effective optimization framework is introduced to automatically identify the optimal teacher-assistant that leads to the maximal student performance. We conduct extensive experiments on multiple image classification datasets, including CIFAR-10, CIFAR-100, and ImageNet. The findings consistently reveal that our approach outperforms several established baselines, paving a path for future knowledge distillation methods on large-scale vision models.

Runjia Zeng, Guangyan Sun, Qifan Wang et al.

Considering deep neural networks as manifold mappers, the pretrain-then-fine-tune paradigm can be interpreted as a two-stage process: pretrain establishes a broad knowledge base, and fine-tune adjusts the model parameters to activate specific neural pathways to align with the target manifold. Although prior fine-tuning approaches demonstrate success, their rigid parameter space limits their ability to dynamically activate appropriate neural pathways, rendering them ill-equipped to adapt flexibly to the diverse and evolving data distributions. In light of this view, we propose a novel approach, Mixture of Expert Prompt Tuning (MEPT), as an effective and efficient manifold-mapping framework. MEPT leverages the Mixture of Experts architecture by integrating multiple prompt experts to adaptively learn diverse and non-stationary data distributions. Empirical evaluations demonstrate that MEPT outperforms several state-of-the-art parameter efficient baselines on SuperGLUE, achieving notable improvements in mean accuracy (e.g., 1.94%) while significantly reducing activated prompts by 79.25%. The effectiveness of MEPT is further supported by theoretical insights from manifold learning and validated through neural activation pathway visualization results. Our code is avaliable at https://runjia.tech/emnlp_mept/.

Zifan Yu, Erfan Bank Tavakoli, Meida Chen et al.

The area of Video Camouflaged Object Detection (VCOD) presents unique challenges in the field of computer vision due to texture similarities between target objects and their surroundings, as well as irregular motion patterns caused by both objects and camera movement. In this paper, we introduce TokenMotion (TMNet), which employs a transformer-based model to enhance VCOD by extracting motion-guided features using a learnable token selection. Evaluated on the challenging MoCA-Mask dataset, TMNet achieves state-of-the-art performance in VCOD. It outperforms the existing state-of-the-art method by a 12.8% improvement in weighted F-measure, an 8.4% enhancement in S-measure, and a 10.7% boost in mean IoU. The results demonstrate the benefits of utilizing motion-guided features via learnable token selection within a transformer-based framework to tackle the intricate task of VCOD.

Prasanna Reddy Pulakurthi, Majid Rabbani, Celso M. de Melo et al.

This paper introduces a novel dual-region augmentation approach designed to reduce reliance on large-scale labeled datasets while improving model robustness and adaptability across diverse computer vision tasks, including source-free domain adaptation (SFDA) and person re-identification (ReID). Our method performs targeted data transformations by applying random noise perturbations to foreground objects and spatially shuffling background patches. This effectively increases the diversity of the training data, improving model robustness and generalization. Evaluations on the PACS dataset for SFDA demonstrate that our augmentation strategy consistently outperforms existing methods, achieving significant accuracy improvements in both single-target and multi-target adaptation settings. By augmenting training data through structured transformations, our method enables model generalization across domains, providing a scalable solution for reducing reliance on manually annotated datasets. Furthermore, experiments on Market-1501 and DukeMTMC-reID datasets validate the effectiveness of our approach for person ReID, surpassing traditional augmentation techniques. The code is available at https://github.com/PrasannaPulakurthi/Foreground-Background-Augmentation

Prasanna Reddy Pulakurthi, Jiamian Wang, Majid Rabbani et al.

Prevalent text-to-video retrieval systems mainly adopt embedding models for feature extraction and compute cosine similarities for ranking. However, this design presents two limitations. Low-quality text-video data pairs could compromise the retrieval, yet are hard to identify and examine. Cosine similarity alone provides no explanation for the ranking results, limiting the interpretability. We ask that can we interpret the ranking results, so as to assess the retrieval models and examine the text-video data? This work proposes X-CoT, an explainable retrieval framework upon LLM CoT reasoning in place of the embedding model-based similarity ranking. We first expand the existing benchmarks with additional video annotations to support semantic understanding and reduce data bias. We also devise a retrieval CoT consisting of pairwise comparison steps, yielding detailed reasoning and complete ranking. X-CoT empirically improves the retrieval performance and produces detailed rationales. It also facilitates the model behavior and data quality analysis. Code and data are available at: https://github.com/PrasannaPulakurthi/X-CoT.

Prasanna Reddy Pulakurthi, Majid Rabbani, Jamison Heard et al.

This work investigates Source-Free Domain Adaptation (SFDA), where a model adapts to a target domain without access to source data. A new augmentation technique, Shuffle PatchMix (SPM), and a novel reweighting strategy are introduced to enhance performance. SPM shuffles and blends image patches to generate diverse and challenging augmentations, while the reweighting strategy prioritizes reliable pseudo-labels to mitigate label noise. These techniques are particularly effective on smaller datasets like PACS, where overfitting and pseudo-label noise pose greater risks. State-of-the-art results are achieved on three major benchmarks: PACS, VisDA-C, and DomainNet-126. Notably, on PACS, improvements of 7.3% (79.4% to 86.7%) and 7.2% are observed in single-target and multi-target settings, respectively, while gains of 2.8% and 0.7% are attained on DomainNet-126 and VisDA-C. This combination of advanced augmentation and robust pseudo-label reweighting establishes a new benchmark for SFDA. The code is available at: https://github.com/PrasannaPulakurthi/SPM

Jiamian Wang, Guohao Sun, Pichao Wang et al.

The increasing prevalence of video clips has sparked growing interest in text-video retrieval. Recent advances focus on establishing a joint embedding space for text and video, relying on consistent embedding representations to compute similarity. However, the text content in existing datasets is generally short and concise, making it hard to fully describe the redundant semantics of a video. Correspondingly, a single text embedding may be less expressive to capture the video embedding and empower the retrieval. In this study, we propose a new stochastic text modeling method T-MASS, i.e., text is modeled as a stochastic embedding, to enrich text embedding with a flexible and resilient semantic range, yielding a text mass. To be specific, we introduce a similarity-aware radius module to adapt the scale of the text mass upon the given text-video pairs. Plus, we design and develop a support text regularization to further control the text mass during the training. The inference pipeline is also tailored to fully exploit the text mass for accurate retrieval. Empirical evidence suggests that T-MASS not only effectively attracts relevant text-video pairs while distancing irrelevant ones, but also enables the determination of precise text embeddings for relevant pairs. Our experimental results show a substantial improvement of T-MASS over baseline (3% to 6.3% by R@1). Also, T-MASS achieves state-of-the-art performance on five benchmark datasets, including MSRVTT, LSMDC, DiDeMo, VATEX, and Charades.

Shoaib Meraj Sami, Md Mahedi Hasan, Nasser M. Nasrabadi et al.

Annotating automatic target recognition (ATR) is a highly challenging task, primarily due to the unavailability of labeled data in the target domain. Hence, it is essential to construct an optimal target domain classifier by utilizing the labeled information of the source domain images. The transductive transfer learning (TTL) method that incorporates a CycleGAN-based unpaired domain translation network has been previously proposed in the literature for effective ATR annotation. Although this method demonstrates great potential for ATR, it severely suffers from lower annotation performance, higher Fréchet Inception Distance (FID) score, and the presence of visual artifacts in the synthetic images. To address these issues, we propose a hybrid contrastive learning base unpaired domain translation (H-CUT) network that achieves a significantly lower FID score. It incorporates both attention and entropy to emphasize the domain-specific region, a noisy feature mixup module to generate high variational synthetic negative patches, and a modulated noise contrastive estimation (MoNCE) loss to reweight all negative patches using optimal transport for better performance. Our proposed contrastive learning and cycle-consistency-based TTL (C3TTL) framework consists of two H-CUT networks and two classifiers. It simultaneously optimizes cycle-consistency, MoNCE, and identity losses. In C3TTL, two H-CUT networks have been employed through a bijection mapping to feed the reconstructed source domain images into a pretrained classifier to guide the optimal target domain classifier. Extensive experimental analysis conducted on three ATR datasets demonstrates that the proposed C3TTL method is effective in annotating civilian and military vehicles, as well as ship targets.

Nitish Mital, Simon Malzard, Richard Walters et al.

Limited real-world data severely impacts model performance in many computer vision domains, particularly for samples that are underrepresented in training. Synthetically generated images are a promising solution, but 1) it remains unclear how to design synthetic training data to optimally improve model performance (e.g, whether and where to introduce more realism or more abstraction) and 2) the domain expertise, time and effort required from human operators for this design and optimisation process represents a major practical challenge. Here we propose a novel conceptual approach to improve the efficiency of designing synthetic images, by using robust Explainable AI (XAI) techniques to guide a human-in-the-loop process of modifying 3D mesh models used to generate these images. Importantly, this framework allows both modifications that increase and decrease realism in synthetic data, which can both improve model performance. We illustrate this concept using a real-world example where data are sparse; detection of vehicles in infrared imagery. We fine-tune an initial YOLOv8 model on the ATR DSIAC infrared dataset and synthetic images generated from 3D mesh models in the Unity gaming engine, and then use XAI saliency maps to guide modification of our Unity models. We show that synthetic data can improve detection of vehicles in orientations unseen in training by 4.6% (to mAP50 = 94.6%). We further improve performance by an additional 1.5% (to 96.1%) through our new XAI-guided approach, which reduces misclassifications through both increasing and decreasing the realism of different parts of the synthetic data. Our proof-of-concept results pave the way for fine, XAI-controlled curation of synthetic datasets tailored to improve object detection performance, whilst simultaneously reducing the burden on human operators in designing and optimising these datasets.

Shoaib Meraj Sami, Md Mahedi Hasan, Nasser M. Nasrabadi et al.

In automatic target recognition (ATR) systems, sensors may fail to capture discriminative, fine-grained detail features due to environmental conditions, noise created by CMOS chips, occlusion, parallaxes, and sensor misalignment. Therefore, multi-sensor image fusion is an effective choice to overcome these constraints. However, multi-modal image sensors are heterogeneous and have domain and granularity gaps. In addition, the multi-sensor images can be misaligned due to intricate background clutters, fluctuating illumination conditions, and uncontrolled sensor settings. In this paper, to overcome these issues, we decompose, align, and fuse multiple image sensor data for target classification. We extract the domain-specific and domain-invariant features from each sensor data. We propose to develop a shared unified discrete token (UDT) space between sensors to reduce the domain and granularity gaps. Additionally, we develop an alignment module to overcome the misalignment between multi-sensors and emphasize the discriminative representation of the UDT space. In the alignment module, we introduce sparsity constraints to provide a better cross-modal representation of the UDT space and robustness against various sensor settings. We achieve superior classification performance compared to single-modality classifiers and several state-of-the-art multi-modal fusion algorithms on four multi-sensor ATR datasets.

Simon Malzard, Nitish Mital, Richard Walters et al.

Computer vision (CV) models for detection, prediction or classification tasks operate on video data-streams that are often degraded in the real world, due to deployment in real-time or on resource-constrained hardware. It is therefore critical that these models are robust to degraded data, but state of the art (SoTA) models are often insufficiently assessed with these real-world constraints in mind. This is exemplified by Skeletal Human Action Recognition (SHAR), which is critical in many CV pipelines operating in real-time and at the edge, but robustness to degraded data has previously only been shallowly and inconsistently assessed. Here we address this issue for SHAR by providing an important first data degradation benchmark on the most detailed and largest 3D open dataset, NTU-RGB+D-120, and assess the robustness of five leading SHAR models to three forms of degradation that represent real-world issues. We demonstrate the need for this benchmark by showing that the form of degradation, which has not previously been considered, has a large impact on model accuracy; at the same effective frame rate, model accuracy can vary by >40% depending on degradation type. We also identify that temporal regularity of frames in degraded SHAR data is likely a major driver of differences in model performance, and harness this to improve performance of existing models by up to >40%, through employing a simple mitigation approach based on interpolation. Finally, we highlight how our benchmark has helped identify an important degradation-resistant SHAR model based in Rough Path Theory; the LogSigRNN SHAR model outperforms the SoTA DeGCN model in five out of six cases at low frame rates by an average accuracy of 6%, despite trailing the SoTA model by 11-12% on un-degraded data at high frame rates (30 FPS).

Cheng Han, Yawen Lu, Guohao Sun et al.

In this work, we introduce the Prototypical Transformer (ProtoFormer), a general and unified framework that approaches various motion tasks from a prototype perspective. ProtoFormer seamlessly integrates prototype learning with Transformer by thoughtfully considering motion dynamics, introducing two innovative designs. First, Cross-Attention Prototyping discovers prototypes based on signature motion patterns, providing transparency in understanding motion scenes. Second, Latent Synchronization guides feature representation learning via prototypes, effectively mitigating the problem of motion uncertainty. Empirical results demonstrate that our approach achieves competitive performance on popular motion tasks such as optical flow and scene depth. Furthermore, it exhibits generality across various downstream tasks, including object tracking and video stabilization.

Cheng Han, James C. Liang, Qifan Wang et al.

We introduce the novel Diffusion Visual Programmer (DVP), a neuro-symbolic image translation framework. Our proposed DVP seamlessly embeds a condition-flexible diffusion model within the GPT architecture, orchestrating a coherent sequence of visual programs (i.e., computer vision models) for various pro-symbolic steps, which span RoI identification, style transfer, and position manipulation, facilitating transparent and controllable image translation processes. Extensive experiments demonstrate DVP's remarkable performance, surpassing concurrent arts. This success can be attributed to several key features of DVP: First, DVP achieves condition-flexible translation via instance normalization, enabling the model to eliminate sensitivity caused by the manual guidance and optimally focus on textual descriptions for high-quality content generation. Second, the framework enhances in-context reasoning by deciphering intricate high-dimensional concepts in feature spaces into more accessible low-dimensional symbols (e.g., [Prompt], [RoI object]), allowing for localized, context-free editing while maintaining overall coherence. Last but not least, DVP improves systemic controllability and explainability by offering explicit symbolic representations at each programming stage, empowering users to intuitively interpret and modify results. Our research marks a substantial step towards harmonizing artificial image translation processes with cognitive intelligence, promising broader applications.

Shoaib M. Sami, Nasser M. Nasrabadi, Raghuveer Rao

One of the major obstacles in designing an automatic target recognition (ATR) algorithm, is that there are often labeled images in one domain (i.e., infrared source domain) but no annotated images in the other target domains (i.e., visible, SAR, LIDAR). Therefore, automatically annotating these images is essential to build a robust classifier in the target domain based on the labeled images of the source domain. Transductive transfer learning is an effective way to adapt a network to a new target domain by utilizing a pretrained ATR network in the source domain. We propose an unpaired transductive transfer learning framework where a CycleGAN model and a well-trained ATR classifier in the source domain are used to construct an ATR classifier in the target domain without having any labeled data in the target domain. We employ a CycleGAN model to transfer the mid-wave infrared (MWIR) images to visible (VIS) domain images (or visible to MWIR domain). To train the transductive CycleGAN, we optimize a cost function consisting of the adversarial, identity, cycle-consistency, and categorical cross-entropy loss for both the source and target classifiers. In this paper, we perform a detailed experimental analysis on the challenging DSIAC ATR dataset. The dataset consists of ten classes of vehicles at different poses and distances ranging from 1-5 kilometers on both the MWIR and VIS domains. In our experiment, we assume that the images in the VIS domain are the unlabeled target dataset. We first detect and crop the vehicles from the raw images and then project them into a common distance of 2 kilometers. Our proposed transductive CycleGAN achieves 71.56% accuracy in classifying the visible domain vehicles in the DSIAC ATR dataset.

Ruiyuan Lin, Suya You, Raghuveer Rao et al.

The construction of a multilayer perceptron (MLP) as a piecewise low-order polynomial approximator using a signal processing approach is presented in this work. The constructed MLP contains one input, one intermediate and one output layers. Its construction includes the specification of neuron numbers and all filter weights. Through the construction, a one-to-one correspondence between the approximation of an MLP and that of a piecewise low-order polynomial is established. Comparison between piecewise polynomial and MLP approximations is made. Since the approximation capability of piecewise low-order polynomials is well understood, our findings shed light on the universal approximation capability of an MLP.

Ruiyuan Lin, Zhiruo Zhou, Suya You et al.

A closed-form solution exists in two-class linear discriminant analysis (LDA), which discriminates two Gaussian-distributed classes in a multi-dimensional feature space. In this work, we interpret the multilayer perceptron (MLP) as a generalization of a two-class LDA system so that it can handle an input composed by multiple Gaussian modalities belonging to multiple classes. Besides input layer $l_{in}$ and output layer $l_{out}$, the MLP of interest consists of two intermediate layers, $l_1$ and $l_2$. We propose a feedforward design that has three stages: 1) from $l_{in}$ to $l_1$: half-space partitionings accomplished by multiple parallel LDAs, 2) from $l_1$ to $l_2$: subspace isolation where one Gaussian modality is represented by one neuron, 3) from $l_2$ to $l_{out}$: class-wise subspace mergence, where each Gaussian modality is connected to its target class. Through this process, we present an automatic MLP design that can specify the network architecture (i.e., the layer number and the neuron number at a layer) and all filter weights in a feedforward one-pass fashion. This design can be generalized to an arbitrary distribution by leveraging the Gaussian mixture model (GMM). Experiments are conducted to compare the performance of the traditional backpropagation-based MLP (BP-MLP) and the new feedforward MLP (FF-MLP).

Yueru Chen, Mozhdeh Rouhsedaghat, Suya You et al.

The successive subspace learning (SSL) principle was developed and used to design an interpretable learning model, known as the PixelHop method,for image classification in our prior work. Here, we propose an improved PixelHop method and call it PixelHop++. First, to make the PixelHop model size smaller, we decouple a joint spatial-spectral input tensor to multiple spatial tensors (one for each spectral component) under the spatial-spectral separability assumption and perform the Saab transform in a channel-wise manner, called the channel-wise (c/w) Saab transform.Second, by performing this operation from one hop to another successively, we construct a channel-decomposed feature tree whose leaf nodes contain features of one dimension (1D). Third, these 1D features are ranked according to their cross-entropy values, which allows us to select a subset of discriminant features for image classification. In PixelHop++, one can control the learning model size of fine-granularity,offering a flexible tradeoff between the model size and the classification performance. We demonstrate the flexibility of PixelHop++ on MNIST, Fashion MNIST, and CIFAR-10 three datasets.

Tong Wu, Prudhvi Gurram, Raghuveer M. Rao et al.

Representation of human actions as a sequence of human body movements or action attributes enables the development of models for human activity recognition and summarization. We present an extension of the low-rank representation (LRR) model, termed the clustering-aware structure-constrained low-rank representation (CS-LRR) model, for unsupervised learning of human action attributes from video data. Our model is based on the union-of-subspaces (UoS) framework, and integrates spectral clustering into the LRR optimization problem for better subspace clustering results. We lay out an efficient linear alternating direction method to solve the CS-LRR optimization problem. We also introduce a hierarchical subspace clustering approach, termed hierarchical CS-LRR, to learn the attributes without the need for a priori specification of their number. By visualizing and labeling these action attributes, the hierarchical model can be used to semantically summarize long video sequences of human actions at multiple resolutions. A human action or activity can also be uniquely represented as a sequence of transitions from one action attribute to another, which can then be used for human action recognition. We demonstrate the effectiveness of the proposed model for semantic summarization and action recognition through comprehensive experiments on five real-world human action datasets.