Computer Vision and Pattern Recognition (cs.CV)

Abstract: Stickers have become a ubiquitous part of modern-day communication, conveying complex emotions through visual imagery. To facilitate the development of more powerful algorithms for analyzing stickers, we propose a large-scale Chinese sticker dataset, namely Sticker820K, which consists of 820k image-text pairs. Each sticker has rich and high-quality textual annotations, including descriptions, optical characters, emotional labels, and style classifications. Although vision-language tasks in the domain of natural images have been well studied, directly applying the those models, such as CLIP, to sticker data is not an optimal solution due to the discrepant nature between natural and emotive image data. Therefore, we propose StickerCLIP as a benchmark model on the Sticker820K dataset. For the text-to-image retrieval task, our StickerCLIP demonstrates strong superiority over the CLIP, which achieves an absolute gain of 66.0\% in mean recall on the Sticker820K test set. Additionally, we endeavor to extend the recently popularized LLM by means of prompt tuning, integrating its ability for sticker retrieval and allowing users to retrieve stickers through instructions. We validate the feasibility of this method, demonstrating the immense potential of prompt tuning in expanding LLM abilities while not affecting the quality of upstream tasks.

Abstract: Breast ultrasound videos contain richer information than ultrasound images, therefore it is more meaningful to develop video models for this diagnosis task. However, the collection of ultrasound video datasets is much harder. In this paper, we explore the feasibility of enhancing the performance of ultrasound video classification using the static image dataset. To this end, we propose KGA-Net and coherence loss. The KGA-Net adopts both video clips and static images to train the network. The coherence loss uses the feature centers generated by the static images to guide the frame attention in the video model. Our KGA-Net boosts the performance on the public BUSV dataset by a large margin. The visualization results of frame attention prove the explainability of our method. The codes and model weights of our method will be made publicly available.

Abstract: We present a novel approach for the detection of deepfake videos using a pair of vision transformers pre-trained by a self-supervised masked autoencoding setup. Our method consists of two distinct components, one of which focuses on learning spatial information from individual RGB frames of the video, while the other learns temporal consistency information from optical flow fields generated from consecutive frames. Unlike most approaches where pre-training is performed on a generic large corpus of images, we show that by pre-training on smaller face-related datasets, namely Celeb-A (for the spatial learning component) and YouTube Faces (for the temporal learning component), strong results can be obtained. We perform various experiments to evaluate the performance of our method on commonly used datasets namely FaceForensics++ (Low Quality and High Quality, along with a new highly compressed version named Very Low Quality) and Celeb-DFv2 datasets. Our experiments show that our method sets a new state-of-the-art on FaceForensics++ (LQ, HQ, and VLQ), and obtains competitive results on Celeb-DFv2. Moreover, our method outperforms other methods in the area in a cross-dataset setup where we fine-tune our model on FaceForensics++ and test on CelebDFv2, pointing to its strong cross-dataset generalization ability.

Abstract: Deepfake technologies empowered by deep learning are rapidly evolving, creating new security concerns for society. Existing multimodal detection methods usually capture audio-visual inconsistencies to expose Deepfake videos. More seriously, the advanced Deepfake technology realizes the audio-visual calibration of the critical phoneme-viseme regions, achieving a more realistic tampering effect, which brings new challenges. To address this problem, we propose a novel Deepfake detection method to mine the correlation between Non-critical Phonemes and Visemes, termed NPVForensics. Firstly, we propose the Local Feature Aggregation block with Swin Transformer (LFA-ST) to construct non-critical phoneme-viseme and corresponding facial feature streams effectively. Secondly, we design a loss function for the fine-grained motion of the talking face to measure the evolutionary consistency of non-critical phoneme-viseme. Next, we design a phoneme-viseme awareness module for cross-modal feature fusion and representation alignment, so that the modality gap can be reduced and the intrinsic complementarity of the two modalities can be better explored. Finally, a self-supervised pre-training strategy is leveraged to thoroughly learn the audio-visual correspondences in natural videos. In this manner, our model can be easily adapted to the downstream Deepfake datasets with fine-tuning. Extensive experiments on existing benchmarks demonstrate that the proposed approach outperforms state-of-the-art methods.

Abstract: This paper investigates the impact of breast density distribution on the generalization performance of deep-learning models on mammography images using the VinDr-Mammo dataset. We explore the use of domain adaptation techniques, specifically Domain Adaptive Object Detection (DAOD) with the Noise Latent Transferability Exploration (NLTE) framework, to improve model performance across breast densities under noisy labeling circumstances. We propose a robust augmentation framework to bridge the domain gap between the source and target inside a dataset. Our results show that DAOD-based methods, along with the proposed augmentation framework, can improve the generalization performance of deep-learning models (+5% overall mAP improvement approximately in our experimental results compared to commonly used detection models). This paper highlights the importance of domain adaptation techniques in medical imaging, particularly in the context of breast density distribution, which is critical in mammography.

Abstract: Zero-shot detection (ZSD), i.e., detection on classes not seen during training, is essential for real world detection use-cases, but remains a difficult task. Recent research attempts ZSD with detection models that output embeddings instead of direct class labels. To this aim, the output of the detection model must be aligned to a learned embedding space such as CLIP. However, this alignment is hindered by detection data sets which are expensive to produce compared to image classification annotations, and the resulting lack of category diversity in the training data. We address this challenge by leveraging the CLIP embedding space in combination with image labels from ImageNet. Our results show that image labels are able to better align the detector output to the embedding space and thus have a high potential for ZSD. Compared to only training on detection data, we see a significant gain by adding image label data of 3.3 mAP for the 65/15 split on COCO on the unseen classes, i.e., we more than double the gain of related work.

Abstract: In recent years, deep neural networks (DNNs) have been found very successful for multi-label classification (MLC) of remote sensing (RS) images. Self-supervised pre-training combined with fine-tuning on a randomly selected small training set has become a popular approach to minimize annotation efforts of data-demanding DNNs. However, fine-tuning on a small and biased training set may limit model performance. To address this issue, we investigate the effectiveness of the joint use of self-supervised pre-training with active learning (AL). The considered AL strategy aims at guiding the MLC fine-tuning of a self-supervised model by selecting informative training samples to annotate in an iterative manner. Experimental results show the effectiveness of applying AL-guided fine-tuning (particularly for the case where strong class-imbalance is present in MLC problems) compared to the application of fine-tuning using a randomly constructed small training set.

Abstract: Unsupervised hashing has received extensive research focus on the past decade, which typically aims at preserving a predefined metric (i.e. Euclidean metric) in the Hamming space. To this end, the encoding functions of the existing hashing are typically quasi-isometric, which devote to reducing the quantization loss from the target metric space to the discrete Hamming space. However, it is indeed problematic to directly minimize such error, since such mentioned two metric spaces are heterogeneous, and the quasi-isometric mapping is non-linear. The former leads to inconsistent feature distributions, while the latter leads to problematic optimization issues. In this paper, we propose a novel unsupervised hashing method, termed Sparsity-Induced Generative Adversarial Hashing (SiGAH), to encode large-scale high-dimensional features into binary codes, which well solves the two problems through a generative adversarial training framework. Instead of minimizing the quantization loss, our key innovation lies in enforcing the learned Hamming space to have similar data distribution to the target metric space via a generative model. In particular, we formulate a ReLU-based neural network as a generator to output binary codes and an MSE-loss based auto-encoder network as a discriminator, upon which a generative adversarial learning is carried out to train hash functions. Furthermore, to generate the synthetic features from the hash codes, a compressed sensing procedure is introduced into the generative model, which enforces the reconstruction boundary of binary codes to be consistent with that of original features. Finally, such generative adversarial framework can be trained via the Adam optimizer. Experimental results on four benchmarks, i.e., Tiny100K, GIST1M, Deep1M, and MNIST, have shown that the proposed SiGAH has superior performance over the state-of-the-art approaches.

Abstract: The weight-sharing mechanism of convolutional kernels ensures translation-equivariance of convolution neural networks (CNNs). Recently, rotation-equivariance has been investigated. However, research on scale-equivariance or simultaneous scale-rotation-equivariance is insufficient. This study proposes a Lie group-CNN, which can keep scale-rotation-equivariance for image classification tasks. The Lie group-CNN includes a lifting module, a series of group convolution modules, a global pooling layer, and a classification layer. The lifting module transfers the input image from Euclidean space to Lie group space, and the group convolution is parameterized through a fully connected network using Lie-algebra of Lie-group elements as inputs to achieve scale-rotation-equivariance. The Lie group SIM(2) is utilized to establish the Lie group-CNN with scale-rotation-equivariance. Scale-rotation-equivariance of Lie group-CNN is verified and achieves the best recognition accuracy on the blood cell dataset (97.50%) and the HAM10000 dataset (77.90%) superior to Lie algebra convolution network, dilation convolution, spatial transformer network, and scale-equivariant steerable network. In addition, the generalization ability of the Lie group-CNN on SIM(2) on rotation-equivariance is verified on rotated-MNIST and rotated-CIFAR10, and the robustness of the network is verified on SO(2) and SE(2). Therefore, the Lie group-CNN can successfully extract geometric features and performs equivariant recognition on images with rotation and scale transformations.

Abstract: Following the successful debut of polyp detection and characterization, more advanced automation tools are being developed for colonoscopy. The new automation tasks, such as quality metrics or report generation, require understanding of the procedure flow that includes activities, events, anatomical landmarks, etc. In this work we present a method for automatic semantic parsing of colonoscopy videos. The method uses a novel DL multi-label temporal segmentation model trained in supervised and unsupervised regimes. We evaluate the accuracy of the method on a test set of over 300 annotated colonoscopy videos, and use ablation to explore the relative importance of various method's components.

Abstract: The imbalanced distribution of long-tailed data poses a challenge for deep neural networks, as models tend to prioritize correctly classifying head classes over others so that perform poorly on tail classes. The lack of semantics for tail classes is one of the key factors contributing to their low recognition accuracy. To rectify this issue, we propose to augment tail classes by borrowing the diverse semantic information from head classes, referred to as head-to-tail fusion (H2T). We randomly replace a portion of the feature maps of the tail class with those of the head class. The fused feature map can effectively enhance the diversity of tail classes by incorporating features from head classes that are relevant to them. The proposed method is easy to implement due to its additive fusion module, making it highly compatible with existing long-tail recognition methods for further performance boosting. Extensive experiments on various long-tailed benchmarks demonstrate the effectiveness of the proposed H2T. The source code is temporarily available at https://github.com/Keke921/H2T.

Abstract: Despite their success in real data synthesis, diffusion models (DMs) often suffer from slow and costly training and sampling issues, limiting their broader applications. To mitigate this, we propose a Fast Diffusion Model (FDM) which improves the diffusion process of DMs from a stochastic optimization perspective to speed up both training and sampling. Specifically, we first find that the diffusion process of DMs accords with the stochastic optimization process of stochastic gradient descent (SGD) on a stochastic time-variant problem. Note that momentum SGD uses both the current gradient and an extra momentum, achieving more stable and faster convergence. We are inspired to introduce momentum into the diffusion process to accelerate both training and sampling. However, this comes with the challenge of deriving the noise perturbation kernel from the momentum-based diffusion process. To this end, we frame the momentum-based process as a Damped Oscillation system whose critically damped state -- the kernel solution -- avoids oscillation and thus has a faster convergence speed of the diffusion process. Empirical results show that our FDM can be applied to several popular DM frameworks, e.g. VP, VE, and EDM, and reduces their training cost by about 50% with comparable image synthesis performance on CIFAR-10, FFHQ, and AFHQv2 datasets. Moreover, FDM decreases their sampling steps by about $3\times$ to achieve similar performance under the same deterministic samplers. The code is available at https://github.com/sail-sg/FDM.

Abstract: Slot attention has shown remarkable object-centric representation learning performance in computer vision tasks without requiring any supervision. Despite its object-centric binding ability brought by compositional modelling, as a deterministic module, slot attention lacks the ability to generate novel scenes. In this paper, we propose the Slot-VAE, a generative model that integrates slot attention with the hierarchical VAE framework for object-centric structured scene generation. For each image, the model simultaneously infers a global scene representation to capture high-level scene structure and object-centric slot representations to embed individual object components. During generation, slot representations are generated from the global scene representation to ensure coherent scene structures. Our extensive evaluation of the scene generation ability indicates that Slot-VAE outperforms slot representation-based generative baselines in terms of sample quality and scene structure accuracy.

Abstract: With the rapid evolution of AI Generated Content (AIGC), forged images produced through this technology are inherently more deceptive and require less human intervention compared to traditional Computer-generated Graphics (CG). However, owing to the disparities between CG and AIGC, conventional CG detection methods tend to be inadequate in identifying AIGC-produced images. To address this issue, our research concentrates on the text-to-image generation process in AIGC. Initially, we first assemble two text-to-image databases utilizing two distinct AI systems, DALLE2 and DreamStudio. Aiming to holistically capture the inherent anomalies produced by AIGC, we develope a robust dual-stream network comprised of a residual stream and a content stream. The former employs the Spatial Rich Model (SRM) to meticulously extract various texture information from images, while the latter seeks to capture additional forged traces in low frequency, thereby extracting complementary information that the residual stream may overlook. To enhance the information exchange between these two streams, we incorporate a cross multi-head attention mechanism. Numerous comparative experiments are performed on both databases, and the results show that our detection method consistently outperforms traditional CG detection techniques across a range of image resolutions. Moreover, our method exhibits superior performance through a series of robustness tests and cross-database experiments. When applied to widely recognized traditional CG benchmarks such as SPL2018 and DsTok, our approach significantly exceeds the capabilities of other existing methods in the field of CG detection.

Abstract: The popularity of data augmentation techniques in machine learning has increased in recent years, as they enable the creation of new samples from existing datasets. Rotational augmentation, in particular, has shown great promise by revolving images and utilising them as additional data points for training. This research study introduces a new approach to enhance the performance of classification methods where the testing sets were generated employing transformations on every image from the original dataset. Subsequently, ensemble-based systems were implemented to determine the most reliable outcome in each subset acquired from the augmentation phase to get a final prediction for every original image. The findings of this study suggest that rotational augmentation techniques can significantly improve the accuracy of standard classification models; and the selection of a voting scheme can considerably impact the model's performance. Overall, the study found that using an ensemble-based voting system produced more accurate results than simple voting.

Abstract: Neural network pruning is a practical way for reducing the size of trained models and the number of floating-point operations. One way of pruning is to use the relative Hessian trace to calculate sensitivity of each channel, as compared to the more common magnitude pruning approach. However, the stochastic approach used to estimate the Hessian trace needs to iterate over many times before it can converge. This can be time-consuming when used for larger models with many millions of parameters. To address this problem, we modify the existing approach by estimating the Hessian trace using FP16 precision instead of FP32. We test the modified approach (EHAP) on ResNet-32/ResNet-56/WideResNet-28-8 trained on CIFAR10/CIFAR100 image classification tasks and achieve faster computation of the Hessian trace. Specifically, our modified approach can achieve speed ups ranging from 17% to as much as 44% during our experiments on different combinations of model architectures and GPU devices. Our modified approach also takes up around 40% less GPU memory when pruning ResNet-32 and ResNet-56 models, which allows for a larger Hessian batch size to be used for estimating the Hessian trace. Meanwhile, we also present the results of pruning using both FP16 and FP32 Hessian trace calculation and show that there are no noticeable accuracy differences between the two. Overall, it is a simple and effective way to compute the relative Hessian trace faster without sacrificing on pruned model performance. We also present a full pipeline using EHAP and quantization aware training (QAT), using INT8 QAT to compress the network further after pruning. In particular, we use symmetric quantization for the weights and asymmetric quantization for the activations.

Abstract: A new data-driven bilateral generalized two-dimensional quaternion principal component analysis (BiG2DQPCA) is presented to extract the features of matrix samples from both row and column directions. This general framework directly works on the 2D color images without vectorizing and well preserves the spatial and color information, which makes it flexible to fit various real-world applications. A generalized ridge regression model of BiG2DQPCA is firstly proposed with orthogonality constrains on aimed features. Applying the deflation technique and the framework of minorization-maximization, a new quaternion optimization algorithm is proposed to compute the optimal features of BiG2DQPCA and a closed-form solution is obtained at each iteration. A new approach based on BiG2DQPCA is presented for color face recognition and image reconstruction with a new data-driven weighting technique. Sufficient numerical experiments are implemented on practical color face databases and indicate the superiority of BiG2DQPCA over the state-of-the-art methods in terms of recognition accuracies and rates of image reconstruction.

Abstract: Vision Transformers (ViTs) have shown impressive performance in computer vision, but their high computational cost, quadratic in the number of tokens, limits their adoption in computation-constrained applications. However, this large number of tokens may not be necessary, as not all tokens are equally important. In this paper, we investigate token pruning to accelerate inference for object detection and instance segmentation, extending prior works from image classification. Through extensive experiments, we offer four insights for dense tasks: (i) tokens should not be completely pruned and discarded, but rather preserved in the feature maps for later use. (ii) reactivating previously pruned tokens can further enhance model performance. (iii) a dynamic pruning rate based on images is better than a fixed pruning rate. (iv) a lightweight, 2-layer MLP can effectively prune tokens, achieving accuracy comparable with complex gating networks with a simpler design. We evaluate the impact of these design choices on COCO dataset and present a method integrating these insights that outperforms prior art token pruning models, significantly reducing performance drop from ~1.5 mAP to ~0.3 mAP for both boxes and masks. Compared to the dense counterpart that uses all tokens, our method achieves up to 34% faster inference speed for the whole network and 46% for the backbone.

Abstract: We present an effective and efficient approach for low-light image enhancement, named Lookup Table Global Curve Estimation (LUT-GCE). In contrast to existing curve-based methods with pixel-wise adjustment, we propose to estimate a global curve for the entire image that allows corrections for both under- and over-exposure. Specifically, we develop a novel cubic curve formulation for light enhancement, which enables an image-adaptive and pixel-independent curve for the range adjustment of an image. We then propose a global curve estimation network (GCENet), a very light network with only 25.4k parameters. To further speed up the inference speed, a lookup table method is employed for fast retrieval. In addition, a novel histogram smoothness loss is designed to enable zero-shot learning, which is able to improve the contrast of the image and recover clearer details. Quantitative and qualitative results demonstrate the effectiveness of the proposed approach. Furthermore, our approach outperforms the state of the art in terms of inference speed, especially on high-definition images (e.g., 1080p and 4k).

Abstract: In self-driving applications, LiDAR data provides accurate information about distances in 3D but lacks the semantic richness of camera data. Therefore, state-of-the-art methods for perception in urban scenes fuse data from both sensor types. In this work, we introduce a novel self-supervised method to fuse LiDAR and camera data for self-driving applications. We build upon masked autoencoders (MAEs) and train deep learning models to reconstruct masked LiDAR data from fused LiDAR and camera features. In contrast to related methods that use birds-eye-view representations, we fuse features from dense spherical LiDAR projections and features from fish-eye camera crops with a similar field of view. Therefore, we reduce the learned spatial transformations to moderate perspective transformations and do not require additional modules to generate dense LiDAR representations. Code is available at: https://github.com/KIT-MRT/masked-fusion-360

Abstract: Cross-modal alignment plays a crucial role in vision-language pre-training (VLP) models, enabling them to capture meaningful associations across different modalities. For this purpose, inspired by the success of masked language modeling (MLM) tasks in the NLP pre-training area, numerous masked modeling tasks have been proposed for VLP to further promote cross-modal interactions. The core idea of previous masked modeling tasks is to focus on reconstructing the masked tokens based on visible context for learning local-local alignment, i.e., associations between image patches and text tokens. However, most of them pay little attention to the global semantic features generated for the masked data, resulting in a limited cross-modal alignment ability of global representations to local features of the other modality. Therefore, in this paper, we propose a novel Global and Local Semantic Completion Learning (GLSCL) task to facilitate global-local alignment and local-local alignment simultaneously. Specifically, the GLSCL task complements the missing semantics of masked data and recovers global and local features by cross-modal interactions. Our GLSCL consists of masked global semantic completion (MGSC) and masked local token completion (MLTC). MGSC promotes learning more representative global features which have a great impact on the performance of downstream tasks, and MLTC can further enhance accurate comprehension on multimodal data. Moreover, we present a flexible vision encoder, enabling our model to simultaneously perform image-text and video-text multimodal tasks. Experimental results show that our proposed method obtains state-of-the-art performance on various vision-language benchmarks, such as visual question answering, image-text retrieval, and video-text retrieval.

Abstract: Enhancing AI systems to perform tasks following human instructions can significantly boost productivity. In this paper, we present InstructP2P, an end-to-end framework for 3D shape editing on point clouds, guided by high-level textual instructions. InstructP2P extends the capabilities of existing methods by synergizing the strengths of a text-conditioned point cloud diffusion model, Point-E, and powerful language models, enabling color and geometry editing using language instructions. To train InstructP2P, we introduce a new shape editing dataset, constructed by integrating a shape segmentation dataset, off-the-shelf shape programs, and diverse edit instructions generated by a large language model, ChatGPT. Our proposed method allows for editing both color and geometry of specific regions in a single forward pass, while leaving other regions unaffected. In our experiments, InstructP2P shows generalization capabilities, adapting to novel shape categories and instructions, despite being trained on a limited amount of data.

Abstract: Deep learning models often face challenges when handling real-world image corruptions. In response, researchers have developed image corruption datasets to evaluate the performance of deep neural networks in handling such corruptions. However, these datasets have a significant limitation: they do not account for all corruptions encountered in real-life scenarios. To address this gap, we present MUFIA (Multiplicative Filter Attack), an algorithm designed to identify the specific types of corruptions that can cause models to fail. Our algorithm identifies the combination of image frequency components that render a model susceptible to misclassification while preserving the semantic similarity to the original image. We find that even state-of-the-art models trained to be robust against known common corruptions struggle against the low visibility-based corruptions crafted by MUFIA. This highlights the need for more comprehensive approaches to enhance model robustness against a wider range of real-world image corruptions.

Abstract: Clouds in remote sensing images inevitably affect information extraction, which hinder the following analysis of satellite images. Hence, cloud detection is a necessary preprocessing procedure. However, the existing methods have numerous calculations and parameters. In this letter, a lightweight CNN-Transformer network, CD-CTFM, is proposed to solve the problem. CD-CTFM is based on encoder-decoder architecture and incorporates the attention mechanism. In the decoder part, we utilize a lightweight network combing CNN and Transformer as backbone, which is conducive to extract local and global features simultaneously. Moreover, a lightweight feature pyramid module is designed to fuse multiscale features with contextual information. In the decoder part, we integrate a lightweight channel-spatial attention module into each skip connection between encoder and decoder, extracting low-level features while suppressing irrelevant information without introducing many parameters. Finally, the proposed model is evaluated on two cloud datasets, 38-Cloud and MODIS. The results demonstrate that CD-CTFM achieves comparable accuracy as the state-of-art methods. At the same time, CD-CTFM outperforms state-of-art methods in terms of efficiency.

Abstract: Contrastive image-text models such as CLIP form the building blocks of many state-of-the-art systems. While they excel at recognizing common generic concepts, they still struggle on fine-grained entities which are rare, or even absent from the pre-training dataset. Hence, a key ingredient to their success has been the use of large-scale curated pre-training data aiming at expanding the set of concepts that they can memorize during the pre-training stage. In this work, we explore an alternative to encoding fine-grained knowledge directly into the model's parameters: we instead train the model to retrieve this knowledge from an external memory. Specifically, we propose to equip existing vision-text models with the ability to refine their embedding with cross-modal retrieved information from a memory at inference time, which greatly improves their zero-shot predictions. Remarkably, we show that this can be done with a light-weight, single-layer, fusion transformer on top of a frozen CLIP. Our experiments validate that our retrieval-enhanced contrastive (RECO) training improves CLIP performance substantially on several challenging fine-grained tasks: for example +10.9 on Stanford Cars, +10.2 on CUB-2011 and +7.3 on the recent OVEN benchmark.

Abstract: Deep neural networks are vulnerable to adversarial examples. Adversarial training (AT) is an effective defense against adversarial examples. However, AT is prone to overfitting which degrades robustness substantially. Recently, data augmentation (DA) was shown to be effective in mitigating robust overfitting if appropriately designed and optimized for AT. This work proposes a new method to automatically learn online, instance-wise, DA policies to improve robust generalization for AT. A novel policy learning objective, consisting of Vulnerability, Affinity and Diversity, is proposed and shown to be sufficiently effective and efficient to be practical for automatic DA generation during AT. This allows our method to efficiently explore a large search space for a more effective DA policy and evolve the policy as training progresses. Empirically, our method is shown to outperform or match all competitive DA methods across various model architectures (CNNs and ViTs) and datasets (CIFAR10, SVHN and Imagenette). Our DA policy reinforced vanilla AT to surpass several state-of-the-art AT methods (with baseline DA) in terms of both accuracy and robustness. It can also be combined with those advanced AT methods to produce a further boost in robustness.

Abstract: Modern text-to-image synthesis models have achieved an exceptional level of photorealism, generating high-quality images from arbitrary text descriptions. In light of the impressive synthesis ability, several studies have exhibited promising results in exploiting generated data for image recognition. However, directly supplementing data-hungry situations in the real-world (e.g. few-shot or long-tailed scenarios) with existing approaches result in marginal performance gains, as they suffer to thoroughly reflect the distribution of the real data. Through extensive experiments, this paper proposes a new image synthesis pipeline for long-tailed situations using Textual Inversion. The study demonstrates that generated images from textual-inverted text tokens effectively aligns with the real domain, significantly enhancing the recognition ability of a standard ResNet50 backbone. We also show that real-world data imbalance scenarios can be successfully mitigated by filling up the imbalanced data with synthetic images. In conjunction with techniques in the area of long-tailed recognition, our method achieves state-of-the-art results on standard long-tailed benchmarks when trained from scratch.

Abstract: Recently, several multi-modal models have been developed for joint image and language understanding, which have demonstrated impressive chat abilities by utilizing advanced large language models (LLMs). The process of developing such models is straightforward yet effective. It involves pre-training an adaptation module to align the semantics of the vision encoder and language model, followed by fine-tuning on the instruction-following data. However, despite the success of this pipeline in image and language understanding, its effectiveness in joint video and language understanding has not been widely explored. In this paper, we aim to develop a novel multi-modal foundation model capable of perceiving video, image, and language within a general framework. To achieve this goal, we introduce Valley: Video Assistant with Large Language model Enhanced ability. Specifically, our proposed Valley model is designed with a simple projection module that bridges video, image, and language modalities, and is further unified with a multi-lingual LLM. We also collect multi-source vision-text pairs and adopt a spatio-temporal pooling strategy to obtain a unified vision encoding of video and image input for pre-training. Furthermore, we generate multi-task instruction-following video data, including multi-shot captions, long video descriptions, action recognition, causal relationship inference, etc. To obtain the instruction-following data, we design diverse rounds of task-oriented conversations between humans and videos, facilitated by ChatGPT. Qualitative examples demonstrate that our proposed model has the potential to function as a highly effective multilingual video assistant that can make complex video understanding scenarios easy. Code, data, and models will be available at https://github.com/RupertLuo/Valley.

Abstract: Over the past decade, Plug-and-Play (PnP) has become a popular method for reconstructing images using a modular framework consisting of a forward and prior model. The great strength of PnP is that an image denoiser can be used as a prior model while the forward model can be implemented using more traditional physics-based approaches. However, a limitation of PnP is that it reconstructs only a single deterministic image. In this paper, we introduce Generative Plug-and-Play (GPnP), a generalization of PnP to sample from the posterior distribution. As with PnP, GPnP has a modular framework using a physics-based forward model and an image denoising prior model. However, in GPnP these models are extended to become proximal generators, which sample from associated distributions. GPnP applies these proximal generators in alternation to produce samples from the posterior. We present experimental simulations using the well-known BM3D denoiser. Our results demonstrate that the GPnP method is robust, easy to implement, and produces intuitively reasonable samples from the posterior for sparse interpolation and tomographic reconstruction. Code to accompany this paper is available at https://github.com/gbuzzard/generative-pnp-allerton .

Abstract: Dust in the agricultural landscape is a significant challenge and influences, for example, the environmental perception of autonomous agricultural machines. Image enhancement algorithms can be used to reduce dust. However, these require dusty and dust-free images of the same environment for validation. In fact, to date, there is no dataset that we are aware of that addresses this issue. Therefore, we present the agriscapes RB-Dust dataset, which is named after its purpose of reference-based dust removal. It is not possible to take pictures from the cabin during tillage, as this would cause shifts in the images. Because of this, we built a setup from which it is possible to take images from a stationary position close to the passing tractor. The test setup was based on a half-sided gate through which the tractor could drive. The field tests were carried out on a farm in Bavaria, Germany, during tillage. During the field tests, other parameters such as soil moisture and wind speed were controlled, as these significantly affect dust development. We validated our dataset with contrast enhancement and image dehazing algorithms and analyzed the generalizability from recordings from the moving tractor. Finally, we demonstrate the application of dust removal based on a high-level vision task, such as person classification. Our empirical study confirms the validity of RB-Dust for vision-based dust removal in agriculture.

Abstract: In this paper, we present MovieFactory, a powerful framework to generate cinematic-picture (3072$\times$1280), film-style (multi-scene), and multi-modality (sounding) movies on the demand of natural languages. As the first fully automated movie generation model to the best of our knowledge, our approach empowers users to create captivating movies with smooth transitions using simple text inputs, surpassing existing methods that produce soundless videos limited to a single scene of modest quality. To facilitate this distinctive functionality, we leverage ChatGPT to expand user-provided text into detailed sequential scripts for movie generation. Then we bring scripts to life visually and acoustically through vision generation and audio retrieval. To generate videos, we extend the capabilities of a pretrained text-to-image diffusion model through a two-stage process. Firstly, we employ spatial finetuning to bridge the gap between the pretrained image model and the new video dataset. Subsequently, we introduce temporal learning to capture object motion. In terms of audio, we leverage sophisticated retrieval models to select and align audio elements that correspond to the plot and visual content of the movie. Extensive experiments demonstrate that our MovieFactory produces movies with realistic visuals, diverse scenes, and seamlessly fitting audio, offering users a novel and immersive experience. Generated samples can be found in YouTube or Bilibili (1080P).

Abstract: The DEtection TRansformer (DETR) algorithm has received considerable attention in the research community and is gradually emerging as a mainstream approach for object detection and other perception tasks. However, the current field lacks a unified and comprehensive benchmark specifically tailored for DETR-based models. To address this issue, we develop a unified, highly modular, and lightweight codebase called detrex, which supports a majority of the mainstream DETR-based instance recognition algorithms, covering various fundamental tasks, including object detection, segmentation, and pose estimation. We conduct extensive experiments under detrex and perform a comprehensive benchmark for DETR-based models. Moreover, we enhance the performance of detection transformers through the refinement of training hyper-parameters, providing strong baselines for supported algorithms.We hope that detrex could offer research communities a standardized and unified platform to evaluate and compare different DETR-based models while fostering a deeper understanding and driving advancements in DETR-based instance recognition. Our code is available at https://github.com/IDEA-Research/detrex. The project is currently being actively developed. We encourage the community to use detrex codebase for further development and contributions.

Abstract: In this paper, we consider the problem of composed image retrieval (CIR), it aims to train a model that can fuse multi-modal information, e.g., text and images, to accurately retrieve images that match the query, extending the user's expression ability. We make the following contributions: (i) we initiate a scalable pipeline to automatically construct datasets for training CIR model, by simply exploiting a large-scale dataset of image-text pairs, e.g., a subset of LAION-5B; (ii) we introduce a transformer-based adaptive aggregation model, TransAgg, which employs a simple yet efficient fusion mechanism, to adaptively combine information from diverse modalities; (iii) we conduct extensive ablation studies to investigate the usefulness of our proposed data construction procedure, and the effectiveness of core components in TransAgg; (iv) when evaluating on the publicly available benckmarks under the zero-shot scenario, i.e., training on the automatically constructed datasets, then directly conduct inference on target downstream datasets, e.g., CIRR and FashionIQ, our proposed approach either performs on par with or significantly outperforms the existing state-of-the-art (SOTA) models. Project page: https://code-kunkun.github.io/ZS-CIR/

Abstract: Cryogenic electron microscopy (cryo-EM) has transformed structural biology by allowing to reconstruct 3D biomolecular structures up to near-atomic resolution. However, the 3D reconstruction process remains challenging, as the 3D structures may exhibit substantial shape variations, while the 2D image acquisition suffers from a low signal-to-noise ratio, requiring to acquire very large datasets that are time-consuming to process. Current reconstruction methods are precise but computationally expensive, or faster but lack a physically-plausible model of large molecular shape variations. To fill this gap, we propose CryoChains that encodes large deformations of biomolecules via rigid body transformation of their polymer instances (chains), while representing their finer shape variations with the normal mode analysis framework of biophysics. Our synthetic data experiments on the human $\text{GABA}_{\text{B}}$ and heat shock protein show that CryoChains gives a biophysically-grounded quantification of the heterogeneous conformations of biomolecules, while reconstructing their 3D molecular structures at an improved resolution compared to the current fastest, interpretable deep learning method.

Abstract: Evaluating the performance of perception modules in autonomous driving is one of the most critical tasks in developing the complex intelligent system. While module-level unit test metrics adopted from traditional computer vision tasks are feasible to some extent, it remains far less explored to measure the impact of perceptual noise on the driving quality of autonomous vehicles in a consistent and holistic manner. In this work, we propose a principled framework that provides a coherent and systematic understanding of the impact an error in the perception module imposes on an autonomous agent's planning that actually controls the vehicle. Specifically, the planning process is formulated as expected utility maximisation, where all input signals from upstream modules jointly provide a world state description, and the planner strives for the optimal action by maximising the expected utility determined by both world states and actions. We show that, under practical conditions, the objective function can be represented as an inner product between the world state description and the utility function in a Hilbert space. This geometric interpretation enables a novel way to analyse the impact of noise in world state estimation on planning and leads to a universal metric for evaluating perception. The whole framework resembles the idea of transcendental idealism in the classical philosophical literature, which gives the name to our approach.

Abstract: We introduce Cap3D, an automatic approach for generating descriptive text for 3D objects. This approach utilizes pretrained models from image captioning, image-text alignment, and LLM to consolidate captions from multiple views of a 3D asset, completely side-stepping the time-consuming and costly process of manual annotation. We apply Cap3D to the recently introduced large-scale 3D dataset, Objaverse, resulting in 660k 3D-text pairs. Our evaluation, conducted using 41k human annotations from the same dataset, demonstrates that Cap3D surpasses human-authored descriptions in terms of quality, cost, and speed. Through effective prompt engineering, Cap3D rivals human performance in generating geometric descriptions on 17k collected annotations from the ABO dataset. Finally, we finetune Text-to-3D models on Cap3D and human captions, and show Cap3D outperforms; and benchmark the SOTA including Point-E, Shape-E, and DreamFusion.

Abstract: Large text-to-image diffusion models have impressive capabilities in generating photorealistic images from text prompts. How to effectively guide or control these powerful models to perform different downstream tasks becomes an important open problem. To tackle this challenge, we introduce a principled finetuning method -- Orthogonal Finetuning (OFT), for adapting text-to-image diffusion models to downstream tasks. Unlike existing methods, OFT can provably preserve hyperspherical energy which characterizes the pairwise neuron relationship on the unit hypersphere. We find that this property is crucial for preserving the semantic generation ability of text-to-image diffusion models. To improve finetuning stability, we further propose Constrained Orthogonal Finetuning (COFT) which imposes an additional radius constraint to the hypersphere. Specifically, we consider two important finetuning text-to-image tasks: subject-driven generation where the goal is to generate subject-specific images given a few images of a subject and a text prompt, and controllable generation where the goal is to enable the model to take in additional control signals. We empirically show that our OFT framework outperforms existing methods in generation quality and convergence speed.

Abstract: The visual classification performance of vision-language models such as CLIP can benefit from additional semantic knowledge, e.g. via large language models (LLMs) such as GPT-3. Further extending classnames with LLM-generated class descriptors, e.g. ``waffle, \textit{which has a round shape}'', or averaging retrieval scores over multiple such descriptors, has been shown to improve generalization performance. In this work, we study this behavior in detail and propose \texttt{Waffle}CLIP, a framework for zero-shot visual classification which achieves similar performance gains on a large number of visual classification tasks by simply replacing LLM-generated descriptors with random character and word descriptors \textbf{without} querying external models. We extend these results with an extensive experimental study on the impact and shortcomings of additional semantics introduced via LLM-generated descriptors, and showcase how semantic context is better leveraged by automatically querying LLMs for high-level concepts, while jointly resolving potential class name ambiguities. Link to the codebase: https://github.com/ExplainableML/WaffleCLIP.