CVPR 2024 论文列表

Vision-language generative AI has demonstrated remarkable promise for empowering cross-modal scene understanding of autonomous driving and high-definition (HD) map systems. However current benchmark datasets lack multi-modal point cloud image and language data pairs. Recent approaches utilize visual instruction learning and cross-modal prompt engineering to expand vision-language models into this domain. In this paper we propose a new vision-language benchmark that can be used to finetune traffic and HD map domain-specific foundation models. Specifically we annotate and leverage large-scale broad-coverage traffic and map data extracted from huge HD map annotations and use CLIP and LLaMA-2 / Vicuna to finetune a baseline model with instruction-following data. Our experimental results across various algorithms reveal that while visual instruction-tuning large language models (LLMs) can effectively learn meaningful representations from MAPLM-QA there remains significant room for further advancements. To facilitate applying LLMs and multi-modal data into self-driving research we will release our visual-language QA data and the baseline models at GitHub.com/LLVM-AD/MAPLM.

Deep learning has achieved remarkable progress in various applications heightening the importance of safeguarding the intellectual property (IP) of well-trained models. It entails not only authorizing usage but also ensuring the deployment of models in authorized data domains i.e. making models exclusive to certain target domains. Previous methods necessitate concurrent access to source training data and target unauthorized data when performing IP protection making them risky and inefficient for decentralized private data. In this paper we target a practical setting where only a well-trained source model is available and investigate how we can realize IP protection. To achieve this we propose a novel MAsk Pruning (MAP) framework. MAP stems from an intuitive hypothesis i.e. there are target-related parameters in a well-trained model locating and pruning them is the key to IP protection. Technically MAP freezes the source model and learns a target-specific binary mask to prevent unauthorized data usage while minimizing performance degradation on authorized data. Moreover we introduce a new metric aimed at achieving a better balance between source and target performance degradation. To verify the effectiveness and versatility we have evaluated MAP in a variety of scenarios including vanilla source-available practical source-free and challenging data-free. Extensive experiments indicate that MAP yields new state-of-the-art performance.

We consider a critical issue of false negatives in Vision- Language Pre-training (VLP) a challenge that arises from the inherent many-to-many correspondence of image-text pairs in large-scale web-crawled datasets. The presence of false negatives can impede achieving optimal performance and even lead to a significant performance drop. To address this challenge we propose MAFA (MAnaging FAlse nega- tives) which consists of two pivotal components building upon the recently developed GRouped mIni-baTch sampling (GRIT) strategy: 1) an efficient connection mining process that identifies and converts false negatives into positives and 2) label smoothing for the image-text contrastive (ITC) loss. Our comprehensive experiments verify the effectiveness of MAFA across multiple downstream tasks emphasizing the crucial role of addressing false negatives in VLP potentially even surpassing the importance of addressing false posi- tives. In addition the compatibility of MAFA with the recent BLIP-family model is also demonstrated. Code is available at https://github.com/jaeseokbyun/MAFA.

Vision-Language Transformers (VLTs) have shown great success recently but are meanwhile accompanied by heavy computation costs where a major reason can be attributed to the large number of visual and language tokens. Existing token pruning research for compressing VLTs mainly follows a single-modality-based scheme yet ignores the critical role of aligning different modalities for guiding the token pruning process causing the important tokens for one modality to be falsely pruned in another modality branch. Meanwhile existing VLT pruning works also lack the flexibility to dynamically compress each layer based on different input samples. To this end we propose a novel framework named Multimodal Alignment-Guided Dynamic Token Pruning (MADTP) for accelerating various VLTs. Specifically we first introduce a well-designed Multi-modality Alignment Guidance (MAG) module that can align features of the same semantic concept from different modalities to ensure the pruned tokens are less important for all modalities. We further design a novel Dynamic Token Pruning (DTP) module which can adaptively adjust the token compression ratio in each layer based on different input instances. Extensive experiments on various benchmarks demonstrate that MADTP significantly reduces the computational complexity of kinds of multimodal models while preserving competitive performance. Notably when applied to the BLIP model in the NLVR2 dataset MADTP can reduce the GFLOPs by 80% with less than 4% performance degradation.

With the success of large language models (LLMs) integrating the vision model into LLMs to build vision-language foundation models has gained much more interest recently. However existing LLM-based large multimodal models (e.g. Video-LLaMA VideoChat) can only take in a limited number of frames for short video understanding. In this study we mainly focus on designing an efficient and effective model for long-term video understanding. Instead of trying to process more frames simultaneously like most existing work we propose to process videos in an online manner and store past video information in a memory bank. This allows our model to reference historical video content for long-term analysis without exceeding LLMs' context length constraints or GPU memory limits. Our memory bank can be seamlessly integrated into current multimodal LLMs in an off-the-shelf manner. We conduct extensive experiments on various video understanding tasks such as long-video understanding video question answering and video captioning and our model can achieve state-of-the-art performances across multiple datasets.

In this paper we present a tensor decomposition and low-rank recovery approach (LowRankOcc) for vision-based 3D semantic occupancy prediction. Conventional methods model outdoor scenes with fine-grained 3D grids but the sparsity of non-empty voxels introduces considerable spatial redundancy leading to potential overfitting risks. In contrast our approach leverages the intrinsic low-rank property of 3D occupancy data factorizing voxel representations into low-rank components to efficiently mitigate spatial redundancy without sacrificing performance. Specifically we present the Vertical-Horizontal (VH) decomposition block factorizes 3D tensors into vertical vectors and horizontal matrices. With our "decomposition-encoding-recovery" framework we encode 3D contexts with only 1/2D convolutions and poolings and subsequently recover the encoded compact yet informative context features back to voxel representations. Experimental results demonstrate that LowRankOcc achieves state-of-the-art performances in semantic scene completion on the SemanticKITTI dataset and 3D occupancy prediction on the nuScenes dataset.

Low-resource settings are well-established in natural lan- guage processing where many languages lack sufficient data for deep learning at scale. However low-resource problems are under-explored in computer vision. In this paper we address this gap and explore the challenges of low-resource image tasks with vision foundation models. We first collect a benchmark of genuinely low-resource image data covering historic maps circuit diagrams and mechanical drawings. These low-resource settings all share three challenges: data scarcity fine-grained differences and the distribution shift from natural images to the specialized domain of interest. While existing foundation models have shown impressive generalizability we find they cannot transfer well to our low-resource tasks. To begin to tackle the challenges of low-resource vision we introduce one simple baseline per challenge. Specifically we i) enlarge the data space by generative models ii) adopt the best sub-kernels to encode local regions for fine-grained difference discovery and iii) learn attention for specialized domains. Experiments on our three low-resource tasks demonstrate our proposals already provide a better baseline than transfer learning data aug- mentation and fine-grained methods. This highlights the unique characteristics and challenges of low-resource vision for foundation models that warrant further investigation. Project page: https://xiaobai1217.github.io/ Low-Resource-Vision/.

The popularity of large-scale pre-training has promoted the development of medical foundation models. However some studies have shown that although foundation models exhibit strong general feature extraction capabilities their performance on specific tasks is still inferior to task-specific methods. In this paper we explore a new perspective called "Knowledge Decomposition" to improve the performance on specific medical tasks which deconstruct the foundation model into multiple lightweight expert models each dedicated to a particular task with the goal of improving specialization while concurrently mitigating resource expenditure. To accomplish the above objective we design a novel framework named Low-Rank Knowledge Decomposition (LoRKD) which explicitly separates graidents by incorporating low-rank expert modules and the efficient knowledge separation convolution. Extensive experimental results demonstrate that the decomposed models perform well in terms of performance and transferability even surpassing the original foundation models. Source code is available at: https://github.com/MediaBrain-SJTU/LoRKD

This paper focuses on the high computational complexity in Large Language Models (LLMs) a significant challenge in both natural language processing (NLP) and multi-modal tasks. We propose Low-Rank Approximation for Sparse At- tention (LoRA-Sparse) an innovative approach that strate- gically reduces this complexity. LoRA-Sparse introduces low-rank linear projection layers for sparse attention ap- proximation. It utilizes an order-mimic training methodol- ogy which is crucial for efficiently approximating the self- attention mechanism in LLMs. We empirically show that sparse attention not only reduces computational demands but also enhances model performance in both NLP and multi-modal tasks. This surprisingly shows that redundant attention in LLMs might be non-beneficial. We extensively validate LoRA-Sparse through rigorous empirical studies in both (NLP) and multi-modal tasks demonstrating its effec- tiveness and general applicability. Based on LLaMA and LLaVA models our methods can reduce more than half of the self-attention computation with even better performance than full-attention baselines.

Vision-and-language navigation (VLN) enables the agent to navigate to a remote location following the natural language instruction in 3D environments. At each navigation step the agent selects from possible candidate locations and then makes the move. For better navigation planning the lookahead exploration strategy aims to effectively evaluate the agent's next action by accurately anticipating the future environment of candidate locations. To this end some existing works predict RGB images for future environments while this strategy suffers from image distortion and high computational cost. To address these issues we propose the pre-trained hierarchical neural radiance representation model (HNR) to produce multi-level semantic features for future environments which are more robust and efficient than pixel-wise RGB reconstruction. Furthermore with the predicted future environmental representations our lookahead VLN model is able to construct the navigable future path tree and select the optimal path via efficient parallel evaluation. Extensive experiments on the VLN-CE datasets confirm the effectiveness of our method.

Look-Up Table (LUT) has recently gained increasing attention for restoring High-Quality (HQ) images from Low-Quality (LQ) observations thanks to its high computational efficiency achieved through a "space for time" strategy of caching learned LQ-HQ pairs. However incorporating multiple LUTs for improved performance comes at the cost of a rapidly growing storage size which is ultimately restricted by the allocatable on-device cache size. In this work we propose a novel LUT compression framework to achieve a better trade-off between storage size and performance for LUT-based image restoration models. Based on the observation that most cached LQ image patches are distributed along the diagonal of a LUT we devise a Diagonal-First Compression (DFC) framework where diagonal LQ-HQ pairs are preserved and carefully re-indexed to maintain the representation capacity while non-diagonal pairs are aggressively subsampled to save storage. Extensive experiments on representative image restoration tasks demonstrate that our DFC framework significantly reduces the storage size of LUT-based models (including our new design) while maintaining their performance. For instance DFC saves up to 90% of storage at a negligible performance drop for x4 super-resolution. The source code is available on GitHub: https://github.com/leenas233/DFC.

The design of deep network architectures and training methods in computer vision has been well-explored. However in almost all cases the images have been used as provided with little exploration of pre-processing steps beyond normalization and data augmentation. Virtually all images posted on the web or captured by devices are processed for viewing by humans. Is the pipeline used for humans also best for use by computers and deep networks? The human visual system uses logarithmic sensors; differences and sums correspond to ratios and products. Features in log space will be invariant to intensity changes and robust to color balance changes. Log RGB space also reveals structure that is corrupted by typical pre-processing. We explore using linear and log RGB data for training standard backbone architectures on an image classification task using data derived directly from RAW images to guarantee its integrity. We found that networks trained on log RGB data exhibit improved performance on an unmodified test set and invariance to intensity and color balance modifications without additional training or data augmentation. Furthermore we found that the gains from using high quality log data could also be partially or fully realized from data in 8-bit sRGB-JPG format by inverting the sRGB transform and taking the log. These results imply existing databases may benefit from this type of pre-processing. While working with log data we found it was critical to retain the integrity of the log relationships and that networks using log data train best with meta-parameters different than those used for sRGB or linear data. Finally we introduce a new 10-category 10k RAW image data set (RAW10) for image classification and other purposes to enable further the exploration of log RGB as an input format for deep networks in computer vision.

The ability to learn from context with novel concepts and deliver appropriate responses are essential in human conversations. Despite current Multimodal Large Language Models (MLLMs) and Large Language Models (LLMs) being trained on mega-scale datasets recognizing unseen images or understanding novel concepts in a training-free manner remains a challenge. In-Context Learning (ICL) explores training-free few-shot learning where models are encouraged to "learn to learn" from limited tasks and generalize to unseen tasks. In this work we propose link-context learning (LCL) which emphasizes "reasoning from cause and effect" to augment the learning capabilities of MLLMs. LCL goes beyond traditional ICL by explicitly strengthening the causal relationship between the support set and the query set. By providing demonstrations with causal links LCL guides the model to discern not only the analogy but also the underlying causal associations between data points which empowers MLLMs to recognize unseen images and understand novel concepts more effectively. To facilitate the evaluation of this novel approach we introduce the ISEKAI dataset comprising exclusively of unseen generated image-label pairs designed for link-context learning. Extensive experiments show that our LCL-MLLM exhibits strong link-context learning capabilities to novel concepts over vanilla MLLMs.

We present a lightweight solution for estimating spatially-coherent indoor lighting from a single RGB image. Previous methods for estimating illumination using volumetric representations have overlooked the sparse distribution of light sources in space necessitating substantial memory and computational resources for achieving high-quality results. We introduce a unified voxel octree-based illumination estimation framework to produce 3D spatially-coherent lighting. Additionally a differentiable voxel octree cone tracing rendering layer is proposed to eliminate regular volumetric representation throughout the entire process and ensure the retention of features across different frequency domains. This reduction significantly decreases spatial usage and required floating-point operations without substantially compromising precision. Experimental results demonstrate that our approach achieves high-quality coherent estimation with minimal cost compared to previous methods.

In recent years there has been an explosion of 2D vision models for numerous tasks such as semantic segmentation style transfer or scene editing enabled by large-scale 2D image datasets. At the same time there has been renewed interest in 3D scene representations such as neural radiance fields from multi-view images. However the availability of 3D or multiview data is still substantially limited compared to 2D image datasets making extending 2D vision models to 3D data highly desirable but also very challenging. Indeed extending a single 2D vision operator like scene editing to 3D typically requires a highly creative method specialized to that task and often requires per-scene optimization. In this paper we ask the question of whether any 2D vision model can be lifted to make 3D consistent predictions. We answer this question in the affirmative; our new Lift3D method trains to predict unseen views on feature spaces generated by a few visual models (i.e. DINO and CLIP) but then generalizes to novel vision operators and tasks such as style transfer super-resolution open vocabulary segmentation and image colorization; for some of these tasks there is no comparable previous 3D method. In many cases we even outperform state-of-the-art methods specialized for the task in question. Moreover Lift3D is a zero-shot method in the sense that it requires no task-specific training nor scene-specific optimization.

Vision-Language Models (VLMs) such as CLIP are trained on large amounts of image-text pairs resulting in remarkable generalization across several data distributions. However in several cases their expensive training and data collection/curation costs do not justify the end application. This motivates a vendor-client paradigm where a vendor trains a large-scale VLM and grants only input-output access to clients on a pay-per-query basis in a black-box setting. The client aims to minimize inference cost by distilling the VLM to a student model using the limited available task-specific data and further deploying this student model in the downstream application. While naive distillation largely improves the In-Domain (ID) accuracy of the student it fails to transfer the superior out-of-distribution (OOD) generalization of the VLM teacher using the limited available labeled images. To mitigate this we propose Vision-Language to Vision - Align Distill Predict (VL2V-ADiP) which first aligns the vision and language modalities of the teacher model with the vision modality of a pre-trained student model and further distills the aligned VLM representations to the student. This maximally retains the pre-trained features of the student while also incorporating the rich representations of the VLM image encoder and the superior generalization of the text embeddings. The proposed approach achieves state-of-the-art results on the standard Domain Generalization benchmarks in a black-box teacher setting as well as a white-box setting where the weights of the VLM are accessible.

Unprocessed RAW video has shown distinct advantages over sRGB video in video editing and computer vision tasks. However capturing RAW video is challenging due to limitations in bandwidth and storage. Various methods have been proposed to address similar issues in single image RAW capture through de-rendering. These methods utilize both the metadata and the sRGB image to perform sRGB-to-RAW de-rendering and recover high-quality single-frame RAW data. However metadata-based methods always require additional computation for online metadata generation imposing severe burden on mobile camera device for high frame rate RAW video capture. To address this issue we propose a framework that utilizes frame affinity to achieve high-quality sRGB-to-RAW video reconstruction. Our approach consists of two main steps. The first step temporal affinity prior extraction uses motion information between adjacent frames to obtain a reference RAW image. The second step spatial feature fusion and mapping learns a pixel-level mapping function using scene-specific and position-specific features provided by the previous frame. Our method can be easily applied to current mobile camera equipment without complicated adaptations or added burden. To demonstrate the effectiveness of our approach we introduce the first RAW Video De-rendering Benchmark. In this benchmark our method outperforms state-of-the-art RAW image reconstruction methods even without image-level metadata.

In Structure-from-Motion (SfM) the underlying viewgraphs of unordered image collections generally have a highly redundant set of edges that can be sparsified for efficiency without significant loss of reconstruction quality. Often there are also false edges due to incorrect image retrieval and repeated structures (symmetries) that give rise to ghosting and superimposed reconstruction artifacts. We present a unified method to simultaneously sparsify the viewgraph and remove false edges. We propose a scoring mechanism based on camera triplets that identifies edge redundancy as well as false edges. Our edge selection is formulated as an optimization problem which can be provably solved using a simple thresholding scheme. This results in a highly efficient algorithm which can be incorporated as a pre-processing step into any SfM pipeline making it practically usable. We demonstrate the utility of our method on generic and ambiguous datasets that cover the range of small medium and large-scale datasets all with different statistical properties. Sparsification of generic datasets using our method significantly reduces reconstruction time while maintaining the accuracy of the reconstructions as well as removing ghosting artifacts. For ambiguous datasets our method removes false edges thereby avoiding incorrect superimposed reconstructions.

Egocentric videos provide a first-person perspective of the wearer's activities involving simultaneous interactions with multiple objects. In this work we propose the task of weakly-supervised Narration-based Video Object Segmentation (NVOS). Given an egocentric video clip and a narration of the wearer's activities our aim is to segment object instances mentioned in the narration without using any spatial annotations during training. Existing weakly-supervised video object grounding methods typically yield bounding boxes for referred objects. In contrast we propose ROSA a weakly-supervised pixel-level grounding framework learning alignments between referred objects and segmentation mask proposals. Our model harnesses vision-language models pre-trained on image-text pairs to embed region masks and object phrases. During training we combine (a) a video-narration contrastive loss that implicitly supervises the alignment between regions and phrases and (b) a region-phrase contrastive loss based on inferred latent alignments. To address the lack of annotated NVOS datasets in egocentric videos we create a new evaluation benchmark VISOR-NVOS leveraging existing annotations of segmentation masks from VISOR alongside 14.6k newly-collected object-based video clip narrations. Our approach achieves state-of-the-art zero-shot pixel-level grounding performance compared to strong baselines under similar supervision. Additionally we demonstrate generalization capabilities for zero-shot video object grounding on YouCook2 a third-person instructional video dataset.