Computer Vision and Pattern Recognition (cs.CV)

Abstract: Ship orientation angle prediction (SOAP) with optical remote sensing images is an important image processing task, which often relies on deep convolutional neural networks (CNNs) to make accurate predictions. This paper proposes a novel framework to reduce the model sizes and computational costs of SOAP models without harming prediction accuracy. First, a new SOAP model called Mobile-SOAP is designed based on MobileNetV2, achieving state-of-the-art prediction accuracy. Four tiny SOAP models are also created by replacing the convolutional blocks in Mobile-SOAP with four small-scale networks, respectively. Then, to transfer knowledge from Mobile-SOAP to four lightweight models, we propose a novel knowledge distillation (KD) framework termed SOAP-KD consisting of a novel feature-based guidance loss and an optimized synthetic samples-based knowledge transfer mechanism. Lastly, extensive experiments on the FGSC-23 dataset confirm the superiority of Mobile-SOAP over existing models and also demonstrate the effectiveness of SOAP-KD in improving the prediction performance of four specially designed tiny models. Notably, by using SOAP-KD, the test mean absolute error of the ShuffleNetV2x1.0-based model is only 8% higher than that of Mobile-SOAP, but its number of parameters and multiply-accumulate operations (MACs) are respectively 61.6% and 60.8% less.

Abstract: In this technical report, we present our findings from a study conducted on the EPIC-KITCHENS-100 Unsupervised Domain Adaptation task for Action Recognition. Our research focuses on the innovative application of a differentiable logic loss in the training to leverage the co-occurrence relations between verb and noun, as well as the pre-trained Large Language Models (LLMs) to generate the logic rules for the adaptation to unseen action labels. Specifically, the model's predictions are treated as the truth assignment of a co-occurrence logic formula to compute the logic loss, which measures the consistency between the predictions and the logic constraints. By using the verb-noun co-occurrence matrix generated from the dataset, we observe a moderate improvement in model performance compared to our baseline framework. To further enhance the model's adaptability to novel action labels, we experiment with rules generated using GPT-3.5, which leads to a slight decrease in performance. These findings shed light on the potential and challenges of incorporating differentiable logic and LLMs for knowledge extraction in unsupervised domain adaptation for action recognition. Our final submission (entitled `NS-LLM') achieved the first place in terms of top-1 action recognition accuracy.

Abstract: Retinal vessel segmentation is generally grounded in image-based datasets collected with bench-top devices. The static images naturally lose the dynamic characteristics of retina fluctuation, resulting in diminished dataset richness, and the usage of bench-top devices further restricts dataset scalability due to its limited accessibility. Considering these limitations, we introduce the first video-based retinal dataset by employing handheld devices for data acquisition. The dataset comprises 635 smartphone-based fundus videos collected from four different clinics, involving 415 patients from 50 to 75 years old. It delivers comprehensive and precise annotations of retinal structures in both spatial and temporal dimensions, aiming to advance the landscape of vasculature segmentation. Specifically, the dataset provides three levels of spatial annotations: binary vessel masks for overall retinal structure delineation, general vein-artery masks for distinguishing the vein and artery, and fine-grained vein-artery masks for further characterizing the granularities of each artery and vein. In addition, the dataset offers temporal annotations that capture the vessel pulsation characteristics, assisting in detecting ocular diseases that require fine-grained recognition of hemodynamic fluctuation. In application, our dataset exhibits a significant domain shift with respect to data captured by bench-top devices, thus posing great challenges to existing methods. In the experiments, we provide evaluation metrics and benchmark results on our dataset, reflecting both the potential and challenges it offers for vessel segmentation tasks. We hope this challenging dataset would significantly contribute to the development of eye disease diagnosis and early prevention.

Abstract: Deception detection is an interdisciplinary field attracting researchers from psychology, criminology, computer science, and economics. We propose a multimodal approach combining deep learning and discriminative models for automated deception detection. Using video modalities, we employ convolutional end-to-end learning to analyze gaze, head pose, and facial expressions, achieving promising results compared to state-of-the-art methods. Due to limited training data, we also utilize discriminative models for deception detection. Although sequence-to-class approaches are explored, discriminative models outperform them due to data scarcity. Our approach is evaluated on five datasets, including a new Rolling-Dice Experiment motivated by economic factors. Results indicate that facial expressions outperform gaze and head pose, and combining modalities with feature selection enhances detection performance. Differences in expressed features across datasets emphasize the importance of scenario-specific training data and the influence of context on deceptive behavior. Cross-dataset experiments reinforce these findings. Despite the challenges posed by low-stake datasets, including the Rolling-Dice Experiment, deception detection performance exceeds chance levels. Our proposed multimodal approach and comprehensive evaluation shed light on the potential of automating deception detection from video modalities, opening avenues for future research.

Abstract: In this work, the novel Image Transformation Sequence Retrieval (ITSR) task is presented, in which a model must retrieve the sequence of transformations between two given images that act as source and target, respectively. Given certain characteristics of the challenge such as the multiplicity of a correct sequence or the correlation between consecutive steps of the process, we propose a solution to ITSR using a general model-based Reinforcement Learning such as Monte Carlo Tree Search (MCTS), which is combined with a deep neural network. Our experiments provide a benchmark in both synthetic and real domains, where the proposed approach is compared with supervised training. The results report that a model trained with MCTS is able to outperform its supervised counterpart in both the simplest and the most complex cases. Our work draws interesting conclusions about the nature of ITSR and its associated challenges.

Abstract: The automatic classification of 3D medical data is memory-intensive. Also, variations in the number of slices between samples is common. Naive solutions such as subsampling can solve these problems, but at the cost of potentially eliminating relevant diagnosis information. Transformers have shown promising performance for sequential data analysis. However, their application for long-sequences is data, computationally, and memory demanding. In this paper, we propose an end-to-end Transformer-based framework that allows to classify volumetric data of variable length in an efficient fashion. Particularly, by randomizing the input slice-wise resolution during training, we enhance the capacity of the learnable positional embedding assigned to each volume slice. Consequently, the accumulated positional information in each positional embedding can be generalized to the neighbouring slices, even for high resolution volumes at the test time. By doing so, the model will be more robust to variable volume length and amenable to different computational budgets. We evaluated the proposed approach in retinal OCT volume classification and achieved 21.96% average improvement in balanced accuracy on a 9-class diagnostic task, compared to state-of-the-art video transformers. Our findings show that varying the slice-wise resolution of the input during training results in more informative volume representation as compared to training with fixed number of slices per volume. Our code is available at: https://github.com/marziehoghbaie/VLFAT.

Abstract: Deep neural networks face many problems in the field of hyperspectral image classification, lack of effective utilization of spatial spectral information, gradient disappearance and overfitting as the model depth increases. In order to accelerate the deployment of the model on edge devices with strict latency requirements and limited computing power, we introduce a lightweight model based on the improved 3D-Densenet model and designs DGCNet. It improves the disadvantage of group convolution. Referring to the idea of dynamic network, dynamic group convolution(DGC) is designed on 3d convolution kernel. DGC introduces small feature selectors for each grouping to dynamically decide which part of the input channel to connect based on the activations of all input channels. Multiple groups can capture different and complementary visual and semantic features of input images, allowing convolution neural network(CNN) to learn rich features. 3D convolution extracts high-dimensional and redundant hyperspectral data, and there is also a lot of redundant information between convolution kernels. DGC module allows 3D-Densenet to select channel information with richer semantic features and discard inactive regions. The 3D-CNN passing through the DGC module can be regarded as a pruned network. DGC not only allows 3D-CNN to complete sufficient feature extraction, but also takes into account the requirements of speed and calculation amount. The inference speed and accuracy have been improved, with outstanding performance on the IN, Pavia and KSC datasets, ahead of the mainstream hyperspectral image classification methods.

Abstract: In the realm of Tiny AI, we introduce "You Only Look at Interested Cells" (YOLIC), an efficient method for object localization and classification on edge devices. Seamlessly blending the strengths of semantic segmentation and object detection, YOLIC offers superior computational efficiency and precision. By adopting Cells of Interest for classification instead of individual pixels, YOLIC encapsulates relevant information, reduces computational load, and enables rough object shape inference. Importantly, the need for bounding box regression is obviated, as YOLIC capitalizes on the predetermined cell configuration that provides information about potential object location, size, and shape. To tackle the issue of single-label classification limitations, a multi-label classification approach is applied to each cell, effectively recognizing overlapping or closely situated objects. This paper presents extensive experiments on multiple datasets, demonstrating that YOLIC achieves detection performance comparable to the state-of-the-art YOLO algorithms while surpassing in speed, exceeding 30fps on a Raspberry Pi 4B CPU. All resources related to this study, including datasets, cell designer, image annotation tool, and source code, have been made publicly available on our project website at https://kai3316.github.io/yolic.github.io

Abstract: We address the video prediction task by putting forth a novel model that combines (i) our recently proposed hierarchical residual vector quantized variational autoencoder (HR-VQVAE), and (ii) a novel spatiotemporal PixelCNN (ST-PixelCNN). We refer to this approach as a sequential hierarchical residual learning vector quantized variational autoencoder (S-HR-VQVAE). By leveraging the intrinsic capabilities of HR-VQVAE at modeling still images with a parsimonious representation, combined with the ST-PixelCNN's ability at handling spatiotemporal information, S-HR-VQVAE can better deal with chief challenges in video prediction. These include learning spatiotemporal information, handling high dimensional data, combating blurry prediction, and implicit modeling of physical characteristics. Extensive experimental results on the KTH Human Action and Moving-MNIST tasks demonstrate that our model compares favorably against top video prediction techniques both in quantitative and qualitative evaluations despite a much smaller model size. Finally, we boost S-HR-VQVAE by proposing a novel training method to jointly estimate the HR-VQVAE and ST-PixelCNN parameters.

Abstract: This paper studies a reconstruction-based approach for weakly-supervised animal detection from aerial images in marine environments. Such an approach leverages an anomaly detection framework that computes metrics directly on the input space, enhancing interpretability and anomaly localization compared to feature embedding methods. Building upon the success of Vector-Quantized Variational Autoencoders in anomaly detection on computer vision datasets, we adapt them to the marine animal detection domain and address the challenge of handling noisy data. To evaluate our approach, we compare it with existing methods in the context of marine animal detection from aerial image data. Experiments conducted on two dedicated datasets demonstrate the superior performance of the proposed method over recent studies in the literature. Our framework offers improved interpretability and localization of anomalies, providing valuable insights for monitoring marine ecosystems and mitigating the impact of human activities on marine animals.

Abstract: Object detection in remote sensing is a crucial computer vision task that has seen significant advancements with deep learning techniques. However, most existing works in this area focus on the use of generic object detection and do not leverage the potential of multimodal data fusion. In this paper, we present a comparison of methods for multimodal object detection in remote sensing, survey available multimodal datasets suitable for evaluation, and discuss future directions.

Abstract: Human Pose Estimation is a thoroughly researched problem; however, most datasets focus on the side and front-view scenarios. We address the limitation by proposing a novel approach that tackles the challenges posed by extreme viewpoints and poses. We introduce a new method for synthetic data generation - RePoGen, RarE POses GENerator - with comprehensive control over pose and view to augment the COCO dataset. Experiments on a new dataset of real images show that adding RePoGen data to the COCO surpasses previous attempts to top-view pose estimation and significantly improves performance on the bottom-view dataset. Through an extensive ablation study on both the top and bottom view data, we elucidate the contributions of methodological choices and demonstrate improved performance. The code and the datasets are available on the project website.

Abstract: Gait Recognition is a computer vision task aiming to identify people by their walking patterns. Existing methods show impressive results on individual datasets but lack the ability to generalize to unseen scenarios. Unsupervised Domain Adaptation (UDA) tries to adapt a model, pre-trained in a supervised manner on a source domain, to an unlabelled target domain. UDA for Gait Recognition is still in its infancy and existing works proposed solutions to limited scenarios. In this paper, we reveal a fundamental phenomenon in adaptation of gait recognition models, in which the target domain is biased to pose-based features rather than identity features, causing a significant performance drop in the identification task. We suggest Gait Orientation-based method for Unsupervised Domain Adaptation (GOUDA) to reduce this bias. To this end, we present a novel Triplet Selection algorithm with a curriculum learning framework, aiming to adapt the embedding space by pushing away samples of similar poses and bringing closer samples of different poses. We provide extensive experiments on four widely-used gait datasets, CASIA-B, OU-MVLP, GREW, and Gait3D, and on three backbones, GaitSet, GaitPart, and GaitGL, showing the superiority of our proposed method over prior works.

Abstract: Vision-language foundation models such as CLIP have shown impressive zero-shot performance on many tasks and datasets, especially thanks to their free-text inputs. However, they struggle to handle some downstream tasks, such as fine-grained attribute detection and localization. In this paper, we propose a multitask fine-tuning strategy based on a positive/negative prompt formulation to further leverage the capacities of the vision-language foundation models. Using the CLIP architecture as baseline, we show strong improvements on bird fine-grained attribute detection and localization tasks, while also increasing the classification performance on the CUB200-2011 dataset. We provide source code for reproducibility purposes: it is available at https://github.com/FactoDeepLearning/MultitaskVLFM.

Abstract: Lane detection plays a pivotal role in the field of autonomous vehicles and advanced driving assistant systems (ADAS). Over the years, numerous algorithms have emerged, spanning from rudimentary image processing techniques to sophisticated deep neural networks. The performance of deep learning-based models is highly dependent on the quality of their training data. Consequently, these models often experience a decline in performance when confronted with challenging scenarios such as extreme lighting conditions, partially visible lane markings, and sparse lane markings like Botts' dots. To address this, we present an end-to-end lane detection and classification system based on deep learning methodologies. In our study, we introduce a unique dataset meticulously curated to encompass scenarios that pose significant challenges for state-of-the-art (SOTA) models. Through fine-tuning selected models, we aim to achieve enhanced localization accuracy. Moreover, we propose a CNN-based classification branch, seamlessly integrated with the detector, facilitating the identification of distinct lane types. This architecture enables informed lane-changing decisions and empowers more resilient ADAS capabilities. We also investigate the effect of using mixed precision training and testing on different models and batch sizes. Experimental evaluations conducted on the widely-used TuSimple dataset, Caltech lane dataset, and our LVLane dataset demonstrate the effectiveness of our model in accurately detecting and classifying lanes amidst challenging scenarios. Our method achieves state-of-the-art classification results on the TuSimple dataset. The code of the work will be published upon the acceptance of the paper.

Abstract: Text-to-image (T2I) personalization allows users to guide the creative image generation process by combining their own visual concepts in natural language prompts. Recently, encoder-based techniques have emerged as a new effective approach for T2I personalization, reducing the need for multiple images and long training times. However, most existing encoders are limited to a single-class domain, which hinders their ability to handle diverse concepts. In this work, we propose a domain-agnostic method that does not require any specialized dataset or prior information about the personalized concepts. We introduce a novel contrastive-based regularization technique to maintain high fidelity to the target concept characteristics while keeping the predicted embeddings close to editable regions of the latent space, by pushing the predicted tokens toward their nearest existing CLIP tokens. Our experimental results demonstrate the effectiveness of our approach and show how the learned tokens are more semantic than tokens predicted by unregularized models. This leads to a better representation that achieves state-of-the-art performance while being more flexible than previous methods.

Abstract: Modular vision-language models (Vision-LLMs) align pretrained image encoders with (pretrained) large language models (LLMs), representing a computationally much more efficient alternative to end-to-end training of large vision-language models from scratch, which is prohibitively expensive for most. Vision-LLMs instead post-hoc condition LLMs to `understand' the output of an image encoder. With the abundance of readily available high-quality English image-text data as well as monolingual English LLMs, the research focus has been on English-only Vision-LLMs. Multilingual vision-language models are still predominantly obtained via expensive end-to-end pretraining, resulting in comparatively smaller models, trained on limited multilingual image data supplemented with text-only multilingual corpora. In this work, we present mBLIP, the first multilingual Vision-LLM, which we obtain in a computationally efficient manner -- on consumer hardware using only a few million training examples -- by leveraging a pretrained multilingual LLM. To this end, we \textit{re-align} an image encoder previously tuned to an English LLM to a new, multilingual LLM -- for this, we leverage multilingual data from a mix of vision-and-language tasks, which we obtain by machine-translating high-quality English data to 95 languages. On the IGLUE benchmark, mBLIP yields results competitive with state-of-the-art models. Moreover, in image captioning on XM3600, mBLIP (zero-shot) even outperforms PaLI-X (a model with 55B parameters). Compared to these very large multilingual vision-language models trained from scratch, we obtain mBLIP by training orders of magnitude fewer parameters on magnitudes less data. We release our model and code at \url{https://github.com/gregor-ge/mBLIP}.

Abstract: Generating videos for visual storytelling can be a tedious and complex process that typically requires either live-action filming or graphics animation rendering. To bypass these challenges, our key idea is to utilize the abundance of existing video clips and synthesize a coherent storytelling video by customizing their appearances. We achieve this by developing a framework comprised of two functional modules: (i) Motion Structure Retrieval, which provides video candidates with desired scene or motion context described by query texts, and (ii) Structure-Guided Text-to-Video Synthesis, which generates plot-aligned videos under the guidance of motion structure and text prompts. For the first module, we leverage an off-the-shelf video retrieval system and extract video depths as motion structure. For the second module, we propose a controllable video generation model that offers flexible controls over structure and characters. The videos are synthesized by following the structural guidance and appearance instruction. To ensure visual consistency across clips, we propose an effective concept personalization approach, which allows the specification of the desired character identities through text prompts. Extensive experiments demonstrate that our approach exhibits significant advantages over various existing baselines.

Abstract: This paper introduces InternVid, a large-scale video-centric multimodal dataset that enables learning powerful and transferable video-text representations for multimodal understanding and generation. The InternVid dataset contains over 7 million videos lasting nearly 760K hours, yielding 234M video clips accompanied by detailed descriptions of total 4.1B words. Our core contribution is to develop a scalable approach to autonomously build a high-quality video-text dataset with large language models (LLM), thereby showcasing its efficacy in learning video-language representation at scale. Specifically, we utilize a multi-scale approach to generate video-related descriptions. Furthermore, we introduce ViCLIP, a video-text representation learning model based on ViT-L. Learned on InternVid via contrastive learning, this model demonstrates leading zero-shot action recognition and competitive video retrieval performance. Beyond basic video understanding tasks like recognition and retrieval, our dataset and model have broad applications. They are particularly beneficial for generating interleaved video-text data for learning a video-centric dialogue system, advancing video-to-text and text-to-video generation research. These proposed resources provide a tool for researchers and practitioners interested in multimodal video understanding and generation.

Abstract: Recent video recognition models utilize Transformer models for long-range spatio-temporal context modeling. Video transformer designs are based on self-attention that can model global context at a high computational cost. In comparison, convolutional designs for videos offer an efficient alternative but lack long-range dependency modeling. Towards achieving the best of both designs, this work proposes Video-FocalNet, an effective and efficient architecture for video recognition that models both local and global contexts. Video-FocalNet is based on a spatio-temporal focal modulation architecture that reverses the interaction and aggregation steps of self-attention for better efficiency. Further, the aggregation step and the interaction step are both implemented using efficient convolution and element-wise multiplication operations that are computationally less expensive than their self-attention counterparts on video representations. We extensively explore the design space of focal modulation-based spatio-temporal context modeling and demonstrate our parallel spatial and temporal encoding design to be the optimal choice. Video-FocalNets perform favorably well against the state-of-the-art transformer-based models for video recognition on three large-scale datasets (Kinetics-400, Kinetics-600, and SS-v2) at a lower computational cost. Our code/models are released at https://github.com/TalalWasim/Video-FocalNets.

Abstract: Prompt learning has emerged as an efficient alternative for fine-tuning foundational models, such as CLIP, for various downstream tasks. Conventionally trained using the task-specific objective, i.e., cross-entropy loss, prompts tend to overfit downstream data distributions and find it challenging to capture task-agnostic general features from the frozen CLIP. This leads to the loss of the model's original generalization capability. To address this issue, our work introduces a self-regularization framework for prompting called PromptSRC (Prompting with Self-regulating Constraints). PromptSRC guides the prompts to optimize for both task-specific and task-agnostic general representations using a three-pronged approach by: (a) regulating {prompted} representations via mutual agreement maximization with the frozen model, (b) regulating with self-ensemble of prompts over the training trajectory to encode their complementary strengths, and (c) regulating with textual diversity to mitigate sample diversity imbalance with the visual branch. To the best of our knowledge, this is the first regularization framework for prompt learning that avoids overfitting by jointly attending to pre-trained model features, the training trajectory during prompting, and the textual diversity. PromptSRC explicitly steers the prompts to learn a representation space that maximizes performance on downstream tasks without compromising CLIP generalization. We perform extensive experiments on 4 benchmarks where PromptSRC overall performs favorably well compared to the existing methods. Our code and pre-trained models are publicly available at: https://github.com/muzairkhattak/PromptSRC.

Abstract: Personalization has emerged as a prominent aspect within the field of generative AI, enabling the synthesis of individuals in diverse contexts and styles, while retaining high-fidelity to their identities. However, the process of personalization presents inherent challenges in terms of time and memory requirements. Fine-tuning each personalized model needs considerable GPU time investment, and storing a personalized model per subject can be demanding in terms of storage capacity. To overcome these challenges, we propose HyperDreamBooth-a hypernetwork capable of efficiently generating a small set of personalized weights from a single image of a person. By composing these weights into the diffusion model, coupled with fast finetuning, HyperDreamBooth can generate a person's face in various contexts and styles, with high subject details while also preserving the model's crucial knowledge of diverse styles and semantic modifications. Our method achieves personalization on faces in roughly 20 seconds, 25x faster than DreamBooth and 125x faster than Textual Inversion, using as few as one reference image, with the same quality and style diversity as DreamBooth. Also our method yields a model that is 10000x smaller than a normal DreamBooth model. Project page: https://hyperdreambooth.github.io