WACV 2023 Papers

This paper introduces pixel-wise prediction based visual odometry(PWVO), which is a dense prediction task that evaluates the values of translation and rotation for every pixel in its input observations. PWVO employs uncertainty estimation to identify the noisy regions in the input observations, and adopts a selection mechanism to integrate pixel-wise predictions based on the estimated uncertainty maps to derive the final translation and rotation. In order to train PWVO in a comprehensive fashion, we further develop a data generation workflow for generating synthetic training data. The experimental results show that PWVO isable to deliver favorable results. In addition, our analyses validate the effectiveness of the designs adopted in PWVO, and demonstrate that the uncertainty mapsestimated by PWVO is capable of capturing the noises in its input observations.

Restoring and inpainting the visual memories that are present, but often impaired, in old photos remains an intriguing but unsolved research topic. Decades-old photos often suffer from severe and commingled degradation such as cracks, defocus, and color-fading, which are difficult to treat individually and harder to repair when they interact. Deep learning presents a plausible avenue, but the lack of large-scale datasets of old photos makes addressing this restoration task very challenging. Here we present a novel reference-based end-to-end learning framework that is able to both repair and colorize old, degraded pictures. Our proposed framework consists of three modules: a restoration sub-network that conducts restoration from degradations, a similarity network that performs color histogram matching and color transfer, and a colorization subnet that learns to predict the chroma elements of images conditioned on chromatic reference signals. The overall system makes uses of color histogram priors from reference images, which greatly reduces the need for large-scale training data. We have also created a first-of-a-kind public dataset of real old photos that are paired with ground truth "pristine" photos that have been manually restored by PhotoShop experts. We conducted extensive experiments on this dataset and synthetic datasets, and found that our method significantly outperforms previous state-of-the-art models using both qualitative comparisons and quantitative measurements. The code is available at https://github.com/DerrickXuNu/Pik-Fix.

We present a new method for generating controllable, dynamically responsive, and photorealistic human animations. Given an image of a person, our system allows the user to generate Physically plausible Upper Body Animation (PUBA) using interaction in the image space, such as dragging their hand to various locations. We formulate a reinforcement learning problem to train a dynamic model that predicts the person's next 2D state (i.e., keypoints on the image) conditioned on a 3D action (i.e., joint torque), and a policy that outputs optimal actions to control the person to achieve desired goals. The dynamic model leverages the expressiveness of 3D simulation and the visual realism of 2D videos. PUBA generates 2D keypoint sequences that achieve task goals while being responsive to forceful perturbation. The sequences of keypoints are then translated by a pose-to-image generator to produce the final photorealistic video.

Adversarial attacks against deep learning-based object detectors have been studied extensively in the past few years. Most of the attacks proposed have targeted the model's integrity (i.e., caused the model to make incorrect predictions), while adversarial attacks targeting the model's availability, a critical aspect in safety-critical domains such as autonomous driving, have not yet been explored by the machine learning research community. In this paper, we propose a novel attack that negatively affects the decision latency of an end-to-end object detection pipeline. We craft a universal adversarial perturbation (UAP) that targets a widely used technique integrated in many object detector pipelines - non-maximum suppression (NMS). Our experiments demonstrate the proposed UAP's ability to increase the processing time of individual frames by adding "phantom" objects that overload the NMS algorithm while preserving the detection of the original objects which allows the attack to go undetected for a longer period of time.

Electrocardiography (ECG), an electrical measurement which captures cardiac activities, is the gold standard for diagnosing cardiovascular disease (CVD). However, ECG is infeasible for continuous cardiac monitoring due to its requirement for user participation. By contrast, photoplethysmography (PPG) provides easy-to-collect data, but its limited accuracy constrains its clinical usage. To combine the advantages of both signals, recent studies incorporate various deep learning techniques for the reconstruction of PPG signals to ECG; however, the lack of contextual information as well as the limited abilities to denoise biomedical signals ultimately constrain model performance. In this research, we propose Performer, a novel Transformer-based architecture that reconstructs ECG from PPG and combines the PPG and reconstructed ECG as multiple modalities for CVD detection. This method is the first time that Transformer sequence-to-sequence translation has been performed on biomedical waveform reconstruction, combining the advantages of both PPG and ECG. We also create Shifted Patch-based Attention (SPA), an effective method to encode/decode the biomedical waveforms. Through fetching the various sequence lengths and capturing cross-patch connections, SPA maximizes the signal processing for both local features and global contextual representations. The proposed architecture generates a state-of-the-art performance of 0.29 RMSE for the reconstruction of PPG to ECG on the BIDMC database, surpassing prior studies. We also evaluated this model on the MIMIC-III dataset, achieving a 95.9% accuracy in CVD detection, and on the PPG-BP dataset, achieving 75.9% accuracy in related CVD diabetes detection, indicating its generalizability. As a proof of concept, an earring wearable named PEARL (prototype), was designed to scale up the point-of-care (POC) healthcare system.

Dynamic Vision Sensors (DVS) are an unconventional type of camera that produces sparse and asynchronous event data, which has recently led to a strong increase in its use for computer vision tasks namely in robotics. Embedded systems face limitations in terms of energy resources, memory, computational power, and communication bandwidth. Hence, this application calls for a way to reduce the amount of data to be processed while keeping the relevant information for the task at hand. We thus believe that a formal definition of event data reduction methods will provide a step further towards sparse data processing. The contributions of this paper are twofold: we introduce two complementary neuromorphic methods based on Spiking Neural Networks for DVS data spatial reduction, which is to best of our knowledge the first proposal of neuromorphic event data reduction; then we study for each method the trade-off between the amount of information kept after reduction, the performance of gesture classification after reduction and their capacity to handle events in real time. We demonstrate here that the proposed SNN-based methods outperform existing methods in a classification task for most dividing factors and are significantly better at handling data in real time, and make therefore the optimal choice for fully-integrated energy-efficient event data reduction running dynamically on a neuromorphic platform. Our code is publicly available online at: https://github.com/amygruel/EvVisu.

Recent advances in camera designs and imaging pipelines allow us to capture high-quality images using smartphones. However, due to the small size and lens limitations of the smartphone cameras, we commonly find artifacts or degradation in the processed images. The most common unpleasant effects are noise artifacts, diffraction artifacts, blur, and HDR overexposure. Deep learning methods for image restoration can successfully remove these artifacts. However, most approaches are not suitable for real-time applications on mobile devices due to their heavy computation and memory requirements. In this paper, we propose LPIENet, a lightweight network for perceptual image enhancement, with the focus on deploying it on smartphones. Our experiments show that, with much fewer parameters and operations, our model can deal with the mentioned artifacts and achieve competitive performance compared with state-of-the-art methods on standard benchmarks. Moreover, to prove the efficiency and reliability of our approach, we deployed the model directly on commercial smartphones and evaluated its performance. Our model can process 2K resolution images under 1s without specific optimization for the mobile devices.

We present Perceiver-VL, a vision-and-language framework that efficiently handles high-dimensional multimodal inputs such as long videos and text. Powered by the iterative latent-cross-attention of Perceiver, our framework scales with linear complexity, in contrast to the quadratic complexity of self-attention used in many state-of-the-art transformer-based models. To further improve the efficiency of our framework, we also study applying LayerDrop on cross-attention layers and introduce a mixed-stream architecture for cross-modal retrieval. We evaluate Perceiver-VL on diverse video-text and image-text benchmarks, where Perceiver-VL achieves the lowest GFLOPs and latency, while maintaining competitive performance. In addition, we also provide comprehensive analyses over various aspects of our framework, including pretraining data, scalability of latent size and input size, dropping cross-attention layers at inference to reduce latency, modality aggregation strategy, positional encoding, and weight initialization strategy.

Adversarial patch attacks mislead neural networks by injecting adversarial pixels within a local region. Patch attacks can be highly effective in a variety of tasks and physically realizable via attachment (e.g. a sticker) to the real-world objects. Despite the diversity in attack patterns, adversarial patches tend to be highly textured and different in appearance from natural images. We exploit this property and present PatchZero, a general defense pipeline against white-box adversarial patches without retraining the downstream classifier or detector. Specifically, our defense detects adversaries at the pixel-level and "zeros out" the patch region by repainting with mean pixel values. We further design a two-stage adversarial training scheme to defend against the stronger adaptive attacks. PatchZero achieves SOTA defense performance on the image classification (ImageNet, RESISC45), object detection (PASCAL VOC), and video classification (UCF101) tasks with little degradation in benign performance. In addition, PatchZero transfers to different patch shapes and attack types.

Vision transformers have demonstrated the potential to outperform CNNs in a variety of vision tasks. But the computational and memory requirements of these models prohibit their use in many applications, especially those that depend on high-resolution images, such as medical image classification. Efforts to train ViTs more efficiently are overly complicated, necessitating architectural changes or intricate training schemes. In this work, we show that standard ViT models can be efficiently trained at high resolution by randomly dropping input image patches. This simple approach, PatchDropout, reduces FLOPs and memory by at least 50% in standard natural image datasets such as ImageNet, and those savings only increase with image size. On CSAW, a high-resolution medical dataset, we observe a 5x savings in computation and memory using PatchDropout, along with a boost in performance. For practitioners with a fixed computational or memory budget, PatchDropout makes it possible to choose image resolution, hyperparameters, or model size to get the most performance out of their model.

Gaze following aims to predict where a person is looking in a scene, by predicting the target location, or indicating that the target is located outside the image. Recent works detect the gaze target by training a heatmap regression task with a pixel-wise mean-square error (MSE) loss, while formulating the in/out prediction task as a binary classification task. This training formulation puts a strict, pixel-level constraint in higher resolution on the single annotation available in training, and does not consider annotation variance and the correlation between the two subtasks. To address these issues, we introduce the patch distribution prediction (PDP) method. We replace the in/out prediction branch in previous models with the PDP branch, by predicting a patch-level gaze distribution that also considers the outside cases. Experiments show that our model regularizes the MSE loss by predicting better heatmap distributions on images with larger annotation variances, meanwhile bridging the gap between the target prediction and in/out prediction subtasks, showing a significant improvement in performance on both subtasks on public gaze following datasets.

As machine learning technology is increasingly adopted into a variety of application domains, the potential risks of data leakage are becoming more serious in certain cases where the data contains highly sensitive information. While some privacy-preserving learning mechanisms for image data, such as SplitNN, enable the training of models without sharing private data on a central server, there exists a trade-off between security and computational cost to a client device. We propose a new mechanism to achieve higher level security and lower computational cost on a client device while maintaining model performance. Our approach, called Patch SplitNN, is based on SplitNN architecture that divides a CNN into two networks, called upper and lower. The difference from that previous work is to input individual image patches into multiple upper models, before concatenating their outputs before the lower model. For further improvement of the upper model training, we introduce an additional network and a loss function into the training process. We demonstrate our Patch SplitNN can classify images as accurately as a ResNet18 on various image classification datasets (CIFAR-10, CIFAR-100, and PCam) under multiple conditions (e.g. patching patterns, dropping patches).

In this paper we study the problem of estimating the semi-generalized pose of a partially calibrated camera, i.e., the pose of a perspective camera with unknown focal length w.r.t. a generalized camera, from a hybrid set of 2D-2D and 2D-3D point correspondences. We study all possible camera configurations within the generalized camera system. To derive practical solvers to previously unsolved challenging configurations, we test different parameterizations as well as different solving strategies based on the state-of-the-art methods for generating efficient polynomial solvers. We evaluate the three most promising solvers, i.e., the H51f solver with five 2D-2D correspondences and one 2D-3D correspondence viewed by the same camera inside generalized camera, the H32f solver with three 2D-2D and two 2D-3D correspondences, and the H13f solver with one 2D-2D and three 2D-3D correspondences, on synthetic and real data. We show that in the presence of noise in the 3D points these solvers provide better estimates than the corresponding absolute pose solvers.

Despite remarkable progress in image translation, the complex scene with multiple discrepant objects remains a challenging problem. The translated images have low fidelity and tiny objects in fewer details causing unsatisfactory performance in object recognition. Without thorough object perception (i.e., bounding boxes, categories, and masks) of images as prior knowledge, the style transformation of each object will be difficult to track in translation. We propose panoptic-aware generative adversarial networks (PanopticGAN) for image-to-image translation together with a compact panoptic segmentation dataset. The panoptic perception (i.e., foreground instances and background semantics of the image scene) is extracted to achieve alignment between object content codes of the input domain and panoptic-level style codes sampled from the target style space, then refined by a proposed feature masking module for sharping object boundaries. The image-level combination between content and sampled style codes is also merged for higher fidelity image generation. Our proposed method was systematically compared with different competing methods and obtained significant improvement in both image quality and object recognition performance.

The extremes of lighting (e.g. too much or too little light) usually cause many troubles for machine and human vision. Many recent works have mainly focused on under-exposure cases where images are often captured in low-light conditions (e.g. nighttime) and achieved promising results for enhancing the quality of images. However, they are inferior to handling images under over-exposure. To mitigate this limitation, we propose a novel unsupervised enhancement framework which is robust against various lighting conditions while does not require any well-exposed images to serve as the ground-truths. Our main concept is to construct pseudo-ground-truth images synthesized from multiple source images that simulate all potential exposure scenarios to train the enhancement network. Our extensive experiments show that the proposed approach consistently outperforms the current state-of-the-art unsupervised counterparts in several public datasets in terms of both quantitative metrics and qualitative results. Our code is available at https://github.com/VinAIResearch/PSENet-Image-Enhancement

Panoptic segmentation is the task of uniquely assigning every pixel in an image to either a semantic label or an individual object instance, generating a coherent and complete scene description. Many current panoptic segmentation methods, however, predict masks of semantic classes and object instances in separate branches, yielding inconsistent predictions. Moreover, because state-of-the-art panoptic segmentation models rely on box proposals, the instance masks predicted are often of low-resolution. To overcome these limitations, we propose the Panoptic Refinement Network (PRN), which takes masks from base panoptic segmentation models and refines them jointly to produce coherent results. PRN extends the offset map-based architecture of Panoptic-Deeplab with several novel ideas including a foreground mask and instance bounding box offsets, as well as coordinate convolutions for improved spatial prediction. Experimental results on COCO and Cityscapes show that PRN can significantly improve already accurate results from a variety of panoptic segmentation networks.

Massive data collected on public roads for autonomous driving has become more popular in many locations in the world. More collected data leads to more concerns about data privacy, including but not limited to pedestrian faces and surrounding vehicle license plates, which urges for robust solutions for detecting and anonymizing them in realistic road-driving scenarios. Existing public datasets for both face and license plate detection are either not focused on autonomous driving or only in parking lots. In this paper, we introduce a challenging public dataset for face and license plate detection in autonomous driving domain. The dataset is aggregated from visual data that is available in public domain, to cover scenarios from six European cities, including daytime and nighttime, annotated with both faces and license plates. All of the images feature a variety of poses and sizes for both faces and license plates. Our dataset offers not only a benchmark for evaluating data anonymization models but also data to get more insights about privacy-preserving autonomous driving. The experimental results showed that 1) current generic state-of-the-art face and/or license plate detection models do not perform well on a realistic and diverse road-driving dataset like ours, 2) our model trained with autonomous driving data (even with soft-labeling data) outperformed strong but generic models, and 3) the size of faces and license plates is an important factor for evaluating and optimizing the performance of privacy-preserving autonomous driving. The annotation of dataset as well as baseline model and results are available at our github: https://github.com/khaclinh/pp4av.

Recently, deep learning has been introduced to solve important medical image reconstruction problems such as sparse-view CT reconstruction. However, the developed deep reconstruction models are generally limited in generalization when applied to unseen testing samples in target domain. Furthermore, privacy concerns may impede the availability of source-domain training data to retrain or adapt the model to the target-domain testing data, which are quite common in real-world medical applications. To address these issues, we introduce a source-free black-box test-time adaptation method for sparse-view CT reconstruction with unknown noise levels based on prior-informed implicit neural representation learning (PINER). By leveraging implicit neural representation learning to generate the image representations at various noise levels, the proposed method is able to construct the adapted input representations at test time based on the inference of black-box model and output analysis. We performed experiments of source-free test-time adaptation for sparse-view CT reconstruction with unknown noise levels on multiple anatomical sites with different black-box deep reconstruction models, where our method outperforms the state-of-the-art algorithms.

The interaction and dimension of points are two important axes in designing point operators to serve hierarchical 3D models. Yet, these two axes are heterogeneous and challenging to fully explore. Existing works craft point operator under a single axis and reuse the crafted operator in all parts of 3D models. This overlooks the opportunity to better combine point interactions and dimensions by exploiting varying geometry/density of 3D point clouds. In this work, we establish PIDS, a novel paradigm to jointly explore point interactions and point dimensions to serve semantic segmentation on point cloud data. We establish a large search space to jointly consider versatile point interactions and point dimensions. This supports point operators with various geometry/density considerations. The enlarged search space with heterogeneous search components calls for a better ranking of candidate models. To achieve this, we improve the search space exploration by leveraging predictor-based Neural Architecture Search (NAS), and enhance the quality of prediction by assigning unique encoding to heterogeneous search components based on their priors. We thoroughly evaluate the networks crafted by PIDS on two semantic segmentation benchmarks, showing 1% mIOU improvement on SemanticKITTI and S3DIS over state-of-the-art 3D models.