WACV 2024 论文列表

Quantizing deep networks with adaptive bit-widths is a promising technique for efficient inference across many devices and resource constraints. In contrast to static methods that repeat the quantization process and train different models for different constraints, adaptive quantization enables us to flexibly adjust the bit-widths of a single deep network during inference for instant adaptation in different scenarios. While existing research shows encouraging results on common image classification benchmarks, this paper investigates how to train such adaptive networks more effectively. Specifically, we present two novel techniques for quantizing deep neural networks with adaptive bit-widths of weights and activations. First, we propose a collaborative strategy to choose a high-precision "teacher" for transferring knowledge to the low-precision "student" while jointly optimizing the model with all bit-widths. Second, to effectively transfer knowledge, we develop a dynamic block swapping method by randomly replacing the blocks in the lower-precision student network with the corresponding blocks in the higher-precision teacher network. Extensive experiments on multiple image classification datasets and novel video classification experiments, well demonstrate the efficacy of our approach over state-of-the-art methods.

Identifying changes in a pair of 3D aerial LiDAR point clouds, obtained during two distinct time periods over the same geographic region presents a significant challenge due to the disparities in spatial coverage and the presence of noise in the acquisition system. The most commonly used approaches to detecting changes in point clouds are based on supervised methods which necessitate extensive labelled data often unavailable in real-world applications. To address these issues, we propose an unsupervised approach that comprises two components: Implicit Neural Representation (INR) for continuous shape reconstruction and a Gaussian Mixture Model for categorising changes. INR offers a grid-agnostic representation for encoding bi-temporal point clouds, with unmatched spatial support that can be regularised to enhance high-frequency details and reduce noise. The reconstructions at each timestamp are compared at arbitrary spatial scales, leading to a significant increase in detection capabilities. We apply our method to a benchmark dataset comprising simulated LiDAR point clouds for urban sprawling. This dataset encompasses diverse challenging scenarios, varying in resolutions, input modalities and noise levels. This enables a comprehensive multi-scenario evaluation, comparing our method with the current state-of-the-art approach. We outperform the previous methods by a margin of 10% in the intersection over union metric. In addition, we put our techniques to practical use by applying them in a real-world scenario to identify instances of illicit excavation of archaeological sites and validate our results by comparing them with findings from field experts.

Existing frameworks for image stitching often provide visually reasonable stitchings. However, they suffer from blurry artifacts and disparities in illumination, depth level, etc. Although the recent learning-based stitchings relax such disparities, the required methods impose sacrifice of image qualities failing to capture high-frequency details for stitched images. To address the problem, we propose a novel approach, implicit Neural Image Stitching (NIS) that extends arbitrary-scale super-resolution. Our method estimates Fourier coefficients of images for quality-enhancing warps. Then, the suggested model blends color mismatches and misalignment in the latent space and decodes the features into RGB values of stitched images. Our experiments show that our approach achieves improvement in resolving the low-definition imaging of the previous deep image stitching with favorable accelerated image-enhancing methods. Our source code is available at https://github.com/minshu-kim/NIS.

Seagrass meadows serve as critical carbon sinks, but estimating the amount of carbon they store requires knowledge of the seagrass species present. Underwater and surface vehicles equipped with machine learning algorithms can help to accurately estimate the composition and extent of seagrass meadows at scale. However, previous approaches for seagrass detection and classification have required supervision from patch-level labels. In this paper, we reframe seagrass classification as a weakly supervised coarse segmentation problem where image-level labels are used during training (25 times fewer labels compared to patch-level labeling) and patch-level outputs are obtained at inference time. To this end, we introduce SeaFeats, an architecture that uses unsupervised contrastive pre-training and feature similarity, and SeaCLIP, a model that showcases the effectiveness of large language models as a supervisory signal in domain-specific applications. We demonstrate that an ensemble of SeaFeats and SeaCLIP leads to highly robust performance. Our method outperforms previous approaches that require patch-level labels on the multi-species 'DeepSeagrass' dataset by 6.8% (absolute) for the class-weighted F1 score, and by 12.1% (absolute) for the seagrass presence/absence F1 score on the 'Global Wetlands' dataset. We also present two case studies for real-world deployment: outlier detection on the Global Wetlands dataset, and application of our method on imagery collected by the FloatyBoat autonomous surface vehicle.

We present a fresh perspective on shot noise corrupted images and noise removal. By viewing image formation as the sequential accumulation of photons on a detector grid, we show that a network trained to predict where the next photon could arrive is in fact solving the minimum mean square error (MMSE) denoising task. This new perspective allows us to make three contributions: (i) We present a new strategy for self-supervised denoising. (ii) We present a new method for sampling from the posterior of possible solutions by iteratively sampling and adding small numbers of photons to the image. (iii) We derive a full generative model by starting this process from an empty canvas. We call this approach generative accumulation of photons (GAP). We evaluate our method quantitatively and qualitatively on 4 new fluorescence microscopy datasets, which will be made available to the community. We find that it outperforms its baselines or performs on-par.

Labeling datasets for supervised object detection is a dull and time-consuming task. Errors can be easily introduced during annotation and overlooked during review, yielding inaccurate benchmarks and performance degradation of deep neural networks trained on noisy labels. In this work, we introduce a benchmark for label error detection methods on object detection datasets as well as a theoretically underpinned label error detection method and a number of baselines. We simulate four different types of randomly introduced label errors on train and test sets of well-labeled object detection datasets. For our label error detection method we assume a two-stage object detector to be given and consider the sum of both stages' classification and regression losses. The losses are computed with respect to the predictions and the noisy labels including simulated label errors, aiming at detecting the latter. We compare our method to four baselines: a naive one without deep learning, the object detector's score, the entropy of the classification softmax distribution and a probability margin based method from related work. We outperform all baselines and demonstrate that among the considered methods, ours is the only one that detects label errors of all four types efficiently, which we also derive theoretically. Furthermore, we detect real label errors a) on commonly used test datasets in object detection and b) on a proprietary dataset. In both cases we achieve low false positives rates, i.e., we detect label errors with a precision for a) of up to 71.5% and for b) with 97%.

Most object-level mapping systems in use today make use of an upstream learned object instance segmentation model. If we want to teach them about a new object or segmentation class, we need to build a large dataset and retrain the system. To build spatial AI systems that can quickly be taught about new objects, we need to effectively solve the problem of single-shot object detection, instance segmentation and re-identification. So far there is neither a method fulfilling all of these requirements in unison nor a benchmark that could be used to test such a method. Addressing this, we propose ISAR, a benchmark and baseline method for single- and few-shot object Instance Segmentation And Re-identification, in an effort to accelerate the development of algorithms that can robustly detect, segment, and re-identify objects from a single or a few sparse training examples. We provide a semi-synthetic dataset of video sequences with ground-truth semantic annotations, a standardized evaluation pipeline, and a baseline method. Our benchmark aligns with the emerging research trend of unifying Multi-Object Tracking, Video Object Segmentation, and Re-identification.

Accelerated magnetic resonance imaging (MRI) aims to reconstruct high-quality MR images from a set of under-sampled measurements. State-of-the-art methods for this task use deep learning, which offers high reconstruction accuracy and fast runtimes. In this work, we propose a new state-of-the-art reconstruction model for accelerated MRI reconstruction. Our model is the first to combine the power of deep neural networks with iterative refinement for this task. For the neural network component of our method, we utilize a transformer-based architecture as transformers are state-of-the-art in various image reconstruction tasks. However, a major drawback of transformers which has limited their emergence among the state-of-the-art MRI models is that they are often memory inefficient for high-resolution inputs. To address this limitation, we propose a transformer-based model which uses parameter-free Fourier-based attention modules, achieving 2x more memory efficiency. We evaluate our model on the largest publicly available MRI dataset, the fastMRI dataset, and achieve on-par performance with other state-of-the-art methods on the dataset's leaderboard.

Implicit Neural Representations (INRs) have revolutionized signal representation by leveraging neural networks to provide continuous and smooth representations of complex data. However, existing INRs face limitations in capturing fine-grained details, handling noise, and adapting to diverse signal types. To address these challenges, we introduce INCODE, a novel approach that enhances the control of the sinusoidal-based activation function in INRs using deep prior knowledge. INCODE comprises a harmonizer network and a composer network, where the harmonizer network dynamically adjusts key parameters of the activation function. Through a task-specific pre-trained model, INCODE adapts the task-specific parameters to optimize the representation process. Our approach not only excels in representation, but also extends its prowess to tackle complex tasks such as audio, image, and 3D shape reconstructions, as well as intricate challenges such as neural radiance fields (NeRFs), and inverse problems, including denoising, super-resolution, inpainting, and CT reconstruction. Through comprehensive experiments, INCODE demonstrates its superiority in terms of robustness, accuracy, quality, and convergence rate, broadening the scope of signal representation.

We propose a novel dataset for ego-view 3D point cloud action recognition. While there has been extensive research on understanding human actions in RGB videos in recent years, the exploration of its 3D point cloud counterpart has been relatively limited. Furthermore, RGB ego-view datasets are rapidly growing, however, 3D point cloud ego-view datasets are scarce at best. Existing 3D datasets are limited in several ways, some include actions that are distinguishable by full-body motion while others use a distant static sensor that hinders the recognition of small objects. We introduce a new point cloud action recognition dataset---the IKEA Ego 3D dataset. It includes sequences of point clouds captured from an ego-view using a HoloLens 2 device. The dataset consists of approximately 493k frames and 56 classes of intricate furniture assembly actions of four different furniture types. We evaluate the performance of various state-of-the-art 3D action recognition methods on the proposed dataset and show that it is very challenging.

Large-scale deployment of fully autonomous vehicles requires a very high degree of robustness to unstructured traffic, weather conditions, and should prevent unsafe mispredictions. While there are several datasets and benchmarks focusing on segmentation for drive scenes, they are not specifically focused on safety and robustness issues. We introduce the IDD-AW dataset, which provides 5000 pairs of high-quality images with pixel-level annotations, captured under rain, fog, low light, and snow in unstructured driving conditions. As compared to other adverse weather datasets, we provide i.) more annotated images, ii.) paired Near-Infrared (NIR) image for each frame, iii.) larger label set with a 4-level label hierarchy to capture unstructured traffic conditions. We benchmark state-of-the-art models for semantic segmentation in IDD-AW. We also propose a new metric called "Safe mean Intersection over Union (Safe mIoU)" for hierarchical datasets which penalizes dangerous mispredictions that are not captured in the traditional definition of mean Intersection over Union (mIoU). The results show that IDD-AW is one of the most challenging datasets to date for these tasks. The dataset and code will be available here: https://iddaw.github.io.

Single image super-resolution (SISR) is a challenging ill-posed problem that aims to up-sample a given low-resolution (LR) image to a high-resolution (HR) counterpart. Due to the difficulty in obtaining real LR-HR training pairs, recent approaches are trained on simulated LR images degraded by simplified down-sampling operators, e.g., bicubic. Such an approach can be problematic in practice due to the large gap between the synthesized and real-world LR images. To alleviate the issue, we propose a novel Invertible scale-Conditional Function (ICF), which can scale an input image and then restore the original input with different scale conditions. Using the proposed ICF, we construct a novel self-supervised SISR framework (ICF-SRSR) to handle the real-world SR task without using any paired/unpaired training data. Furthermore, our ICF-SRSR can generate realistic and feasible LR-HR pairs, which can make existing supervised SISR networks more robust. Extensive experiments demonstrate the effectiveness of our method in handling SISR in a fully self-supervised manner. Our ICF-SRSR demonstrates superior performance compared to the existing methods trained on synthetic paired images in real-world scenarios and exhibits comparable performance compared to state-of-the-art supervised/unsupervised methods on public benchmark datasets. The code is available from this link.

In the field of chest X-ray (CXR) diagnosis, existing works often focus solely on determining where a radiologist looks, typically through tasks such as detection, segmentation, or classification. However, these approaches are often designed as black-box models, lacking interpretability. In this paper, we introduce Interpretable Artificial Intelligence (I-AI) a novel and unified controllable interpretable pipeline for decoding the intense focus of radiologists in CXR diagnosis. Our I-AI addresses three key questions: where a radiologist looks, how long they focus on specific areas, and what findings they diagnose. By capturing the intensity of the radiologist's gaze, we provide a unified solution that offers insights into the cognitive process underlying radiological interpretation. Unlike current methods that rely on black-box machine learning models, which can be prone to extracting erroneous information from the entire input image during the diagnosis process, we tackle this issue by effectively masking out irrelevant information. Our proposed I-AI leverages a vision-language model, allowing for precise control over the interpretation process while ensuring the exclusion of irrelevant features. To train our I-AI model, we utilize an eye gaze dataset to extract anatomical gaze information and generate ground truth heatmaps. Through extensive experimentation, we demonstrate the efficacy of our method. We showcase that the attention heatmaps, designed to mimic radiologists' focus, encode sufficient and relevant information, enabling accurate classification tasks using only a portion of CXR. The code, checkpoints, and data are at https://github.com/UARK-AICV/IAI.

Recent research in representation learning has shown that hierarchical data lends itself to low-dimensional and highly informative representations in hyperbolic space. However, even if hyperbolic embeddings have gathered attention in image recognition, their optimization is prone to numerical hurdles. Further, it remains unclear which applications stand to benefit the most from the implicit bias imposed by hyperbolicity, when compared to traditional Euclidean features. In this paper, we focus on prototypical hyperbolic neural networks. In particular, the tendency of hyperbolic embeddings to converge to the boundary of the Poincare ball in high dimensions and the effect this has on few-shot classification. We show that the best few-shot results are attained for hyperbolic embeddings at a common hyperbolic radius. In contrast to prior benchmark results, we demonstrate that better performance can be achieved by a fixed-radius encoder equipped with the Euclidean metric, regardless of the embedding dimension.

Out-of-distribution (OOD) detection is an important topic for real-world machine learning systems, but settings with limited in-distribution samples have been underexplored. Such few-shot OOD settings are challenging, as models have scarce opportunities to learn the data distribution before being tasked with identifying OOD samples. Indeed, we demonstrate that recent state-of-the-art OOD methods fail to outperform simple baselines in the few-shot setting. We thus propose a hypernetwork framework called HyperMix, using Mixup on the generated classifier parameters, as well as a natural out-of-episode outlier exposure technique that does not require an additional outlier dataset. We conduct experiments on CIFAR-FS and MiniImageNet, significantly outperforming other OOD methods in the few-shot regime.

Deep learning models are known to suffer from the problem of bias, and researchers have been exploring methods to address this issue. However, most of these methods require prior knowledge of the bias and are not always practical. In this paper, we focus on a more practical setting with no prior information about the bias. Generally, in this setting, there are a large number of bias-aligned samples that cause the model to produce biased predictions and a few bias-conflicting samples that do not conform to the bias. If the training data is limited, the influence of the bias-aligned samples may become even stronger on the model predictions, and we experimentally demonstrate that existing debiasing techniques suffer severely in such cases. In this paper, we examine the effects of unknown bias in small dataset regimes and present a novel approach to mitigate this issue. The proposed approach directly addresses the issue of the extremely low occurrence of bias-conflicting samples in limited data settings through the synthesis of hybrid samples that can be used to reduce the effect of bias. We perform extensive experiments on several benchmark datasets and experimentally demonstrate the effectiveness of our proposed approach in addressing any unknown bias in the presence of limited data. Specifically, our approach outperforms the vanilla, LfF, LDD, and DebiAN debiasing methods by absolute margins of 10.39%, 9.08%, 8.07%, and 9.67% when only 10% of the Corrupted CIFAR-10 Type 1 dataset is available with a bias-conflicting sample ratio of 0.05.

Deep Implicit Functions (DIFs) have gained popularity in 3D computer vision due to their compactness and continuous representation capabilities. However, addressing dense correspondences and semantic relationships across DIF-encoded shapes remains a critical challenge, limiting their applications in texture transfer and shape analysis. Moreover, recent endeavors in 3D shape generation using DIFs often neglect correspondence and topology preservation. This paper presents HNDF (Hybrid Neural Diffeomorphic Flow), a method that implicitly learns the underlying representation and decomposes intricate dense correspondences into explicitly axis-aligned triplane features. To avoid suboptimal representations trapped in local minima, we propose hybrid supervision that captures both local and global correspondences. Unlike conventional approaches that directly generate new 3D shapes, we further explore the idea of shape generation with deformed template shape via diffeomorphic flows, where the deformation is encoded by the generated triplane features. Leveraging a pre-existing 2D diffusion model, we produce high-quality and diverse 3D diffeomorphic flows through generated triplanes features, ensuring topological consistency with the template shape. Extensive experiments on medical image organ segmentation datasets evaluate the effectiveness of HNDF in 3D shape representation and generation.

Recent advances in Neural radiance fields (NeRF) have enabled high-fidelity scene reconstruction for novel view synthesis. However, NeRF requires hundreds of network evaluations per pixel to approximate a volume rendering integral, making it slow to train. Caching NeRFs into explicit data structures can effectively enhance rendering speed but at the cost of higher memory usage. To address these issues, we present Hyb-NeRF, a novel neural radiance field with a multi-resolution hybrid encoding that achieves efficient neural modeling and fast rendering, which also allows for high-quality novel view synthesis. The key idea of Hyb-NeRF is to represent the scene using different encoding strategies from coarse-to-fine resolution levels. Hyb-NeRF exploits coherence and compact memory of learnable positional features at coarse resolutions and the fast optimization speed and local details of hash-based feature grids at fine resolutions. In addition, to further boost performance, we embed cone tracing-based Fourier features in our learnable positional encoding that eliminates encoding ambiguity and reduces aliasing artifacts. Extensive experiments on both synthetic and real-world datasets show that Hyb-NeRF achieves faster rendering speed with better rending quality and even a lower memory footprint in comparison to previous state-of-the-art methods.

With the rise in expectations related to generative models, text-to-video (T2V) models are being actively studied. Existing text-to-video models have limitations such as in generating complex movements replicating human motions. These model often generate unintended human motions, and the scale of the subject is incorrect. To overcome these limitations and generate high-quality videos that depict human motion under plausible viewing angles, we propose a two stage framework in this study. In the first stage a text-driven human motion generation network generates three-dimensional (3D) human motion from input text prompts and then motion-to-skeleton projection module projects generated motions onto a two-dimensional (2D) skeleton. In the second stage, the projected skeletons are used to generate a video in which the movements of a subject are well-represented. We demonstrated that the proposed framework quantitatively and qualitatively outperforms the existing T2V models. Previously reported human motion generation models use texts only or texts and human skeletons. However, our framework only uses texts and outputs a video related to human motion. Moreover, our framework benefits from using skeleton as an additional condition in the text-to-human motion generation networks. To the best of our knowledge, our framework is the first of its kind that uses text-driven human motion generation networks to generate high-quality videos related to human motions. The corresponding codes are available at https://github.com/CSJasper/HMTV.