An Empirical Study on the Efficacy of Deep Active Learning for Image Classification

Deep Active Learning (DAL) has been advocated as a promising method to reduce labeling costs in supervised learning. However, existing evaluations of DAL methods are based on different settings, and their results are controversial. To tackle this issue, this paper comprehensively evaluates 19 existing DAL methods in a uniform setting, including traditional fully-\underline{s}upervised \underline{a}ctive \underline{l}earning (SAL) strategies and emerging \underline{s}emi-\underline{s}upervised \underline{a}ctive \underline{l}earning (SSAL) techniques. We have several non-trivial findings. First, most SAL methods cannot achieve higher accuracy than random selection. Second, semi-supervised training brings significant performance improvement compared to pure SAL methods. Third, performing data selection in the SSAL setting can achieve a significant and consistent performance improvement, especially with abundant unlabeled data. Our findings produce the following guidance for practitioners: one should (i) apply SSAL early and (ii) collect more unlabeled data whenever possible, for better model performance.

Learning Single Image Defocus Deblurring with Misaligned Training Pairs

By adopting popular pixel-wise loss, existing methods for defocus deblurring heavily rely on well aligned training image pairs. Although training pairs of ground-truth and blurry images are carefully collected, e.g., DPDD dataset, misalignment is inevitable between training pairs, making existing methods possibly suffer from deformation artifacts. In this paper, we propose a joint deblurring and reblurring learning (JDRL) framework for single image defocus deblurring with misaligned training pairs. Generally, JDRL consists of a deblurring module and a spatially invariant reblurring module, by which deblurred result can be adaptively supervised by ground-truth image to recover sharp textures while maintaining spatial consistency with the blurry image. First, in the deblurring module, a bi-directional optical flow-based deformation is introduced to tolerate spatial misalignment between deblurred and ground-truth images. Second, in the reblurring module, deblurred result is reblurred to be spatially aligned with blurry image, by predicting a set of isotropic blur kernels and weighting maps. Moreover, we establish a new single image defocus deblurring (SDD) dataset, further validating our JDRL and also benefiting future research. Our JDRL can be applied to boost defocus deblurring networks in terms of both quantitative metrics and visual quality on DPDD, RealDOF and our SDD datasets.

Robust Human Matting via Semantic Guidance

Automatic human matting is highly desired for many real applications. We investigate recent human matting methods and show that common bad cases happen when semantic human segmentation fails. This indicates that semantic understanding is crucial for robust human matting. From this, we develop a fast yet accurate human matting framework, named Semantic Guided Human Matting (SGHM). It builds on a semantic human segmentation network and introduces a light-weight matting module with only marginal computational cost. Unlike previous works, our framework is data efficient, which requires a small amount of matting ground-truth to learn to estimate high quality object mattes. Our experiments show that trained with merely 200 matting images, our method can generalize well to real-world datasets, and outperform recent methods on multiple benchmarks, while remaining efficient. Considering the unbearable labeling cost of matting data and widely available segmentation data, our method becomes a practical and effective solution for the task of human matting. Source code is available at https://github.com/cxgincsu/SemanticGuidedHumanMatting.

GTAV-NightRain: Photometric Realistic Large-scale Dataset for Night-time Rain Streak Removal

Rain is transparent, which reflects and refracts light in the scene to the camera. In outdoor vision, rain, especially rain streaks degrade visibility and therefore need to be removed. In existing rain streak removal datasets, although density, scale, direction and intensity have been considered, transparency is not fully taken into account. This problem is particularly serious in night scenes, where the appearance of rain largely depends on the interaction with scene illuminations and changes drastically on different positions within the image. This is problematic, because unrealistic dataset causes serious domain bias. In this paper, we propose GTAV-NightRain dataset, which is a large-scale synthetic night-time rain streak removal dataset. Unlike existing datasets, by using 3D computer graphic platform (namely GTA V), we are allowed to infer the three dimensional interaction between rain and illuminations, which insures the photometric realness. Current release of the dataset contains 12,860 HD rainy images and 1,286 corresponding HD ground truth images in diversified night scenes. A systematic benchmark and analysis are provided along with the dataset to inspire further research.

ViTKD: Practical Guidelines for ViT feature knowledge distillation

Knowledge Distillation (KD) for Convolutional Neural Network (CNN) is extensively studied as a way to boost the performance of a small model. Recently, Vision Transformer (ViT) has achieved great success on many computer vision tasks and KD for ViT is also desired. However, besides the output logit-based KD, other feature-based KD methods for CNNs cannot be directly applied to ViT due to the huge structure gap. In this paper, we explore the way of feature-based distillation for ViT. Based on the nature of feature maps in ViT, we design a series of controlled experiments and derive three practical guidelines for ViT's feature distillation. Some of our findings are even opposite to the practices in the CNN era. Based on the three guidelines, we propose our feature-based method ViTKD which brings consistent and considerable improvement to the student. On ImageNet-1k, we boost DeiT-Tiny from 74.42% to 76.06%, DeiT-Small from 80.55% to 81.95%, and DeiT-Base from 81.76% to 83.46%. Moreover, ViTKD and the logit-based KD method are complementary and can be applied together directly. This combination can further improve the performance of the student. Specifically, the student DeiT-Tiny, Small, and Base achieve 77.78%, 83.59%, and 85.41%, respectively. The code is available at https://github.com/yzd-v/cls_KD.

Rethinking Knowledge Distillation via Cross-Entropy

Knowledge Distillation (KD) has developed extensively and boosted various tasks. The classical KD method adds the KD loss to the original cross-entropy (CE) loss. We try to decompose the KD loss to explore its relation with the CE loss. Surprisingly, we find it can be regarded as a combination of the CE loss and an extra loss which has the identical form as the CE loss. However, we notice the extra loss forces the student's relative probability to learn the teacher's absolute probability. Moreover, the sum of the two probabilities is different, making it hard to optimize. To address this issue, we revise the formulation and propose a distributed loss. In addition, we utilize teachers' target output as the soft target, proposing the soft loss. Combining the soft loss and the distributed loss, we propose a new KD loss (NKD). Furthermore, we smooth students' target output to treat it as the soft target for training without teachers and propose a teacher-free new KD loss (tf-NKD). Our method achieves state-of-the-art performance on CIFAR-100 and ImageNet. For example, with ResNet-34 as the teacher, we boost the ImageNet Top-1 accuracy of ResNet18 from 69.90% to 71.96%. In training without teachers, MobileNet, ResNet-18 and SwinTransformer-Tiny achieve 70.04%, 70.76%, and 81.48%, which are 0.83%, 0.86%, and 0.30% higher than the baseline, respectively. The code is available at https://github.com/yzd-v/cls_KD.

Deep Learning-enabled Spatial Phase Unwrapping for 3D Measurement

In terms of 3D imaging speed and system cost, the single-camera system projecting single-frequency patterns is the ideal option among all proposed Fringe Projection Profilometry (FPP) systems. This system necessitates a robust spatial phase unwrapping (SPU) algorithm. However, robust SPU remains a challenge in complex scenes. Quality-guided SPU algorithms need more efficient ways to identify the unreliable points in phase maps before unwrapping. End-to-end deep learning SPU methods face generality and interpretability problems. This paper proposes a hybrid method combining deep learning and traditional path-following for robust SPU in FPP. This hybrid SPU scheme demonstrates better robustness than traditional quality-guided SPU methods, better interpretability than end-to-end deep learning scheme, and generality on unseen data. Experiments on the real dataset of multiple illumination conditions and multiple FPP systems differing in image resolution, the number of fringes, fringe direction, and optics wavelength verify the effectiveness of the proposed method.

A Simulation Study of Passing Drivers' Responses to the Automated Truck-Mounted Attenuator System in Road Maintenance

The Autonomous Truck-Mounted Attenuator (ATMA) system is a lead-follower vehicle system based on autonomous driving and connected vehicle technologies. The lead truck performs maintenance tasks on the road, and the unmanned follower truck is designed to improve the visibility of the moving work zone to passing vehicles and to protect workers and equipment. While the ATMA has been under testing by transportation maintenance and operations agencies in recent years, a simulator-based testing capability is a supplement, especially if human subjects are involved. This paper aims to discover how passing drivers perceive, understand, and react to the ATMA system in road maintenance accordingly. A driving simulator for ATMA studies is developed for collecting the driving data. Then, driving simulation experiments were performed, wherein a screen-based eye tracker collected sixteen subjects' gaze points and pupil diameters. Data analysis has evidenced the changes in the visual attention pattern of subjects when they were passing the ATMA. On average, the ATMA starts to attract subjects' attention from 500 ft behind the follower truck. Most (87.50%) understood the follower truck's protection purpose, and the majority (86.67%) reasoned the association between the two trucks. But still, many (43.75%) did not recognize that ATMA is a connected autonomous vehicle system. While all subjects safely changed lanes and attempted to pass the slow-moving ATMA, their inadequate understanding of ATMA is a potential risk, like cutting into the ATAM. Results implied that transportation maintenance and operations agencies should take this into consideration in establishing the deployment guidance.

Using Chatbots to Teach Languages

This paper reports on progress towards building an online language learning tool to provide learners with conversational experience by using dialog systems as conversation practice partners. Our system can adapt to users' language proficiency on the fly. We also provide automatic grammar error feedback to help users learn from their mistakes. According to our first adopters, our system is entertaining and useful. Furthermore, we will provide the learning technology community a large-scale conversation dataset on language learning and grammar correction. Our next step is to make our system more adaptive to user profile information by using reinforcement learning algorithms.

A Multi-tasking Model of Speaker-Keyword Classification for Keeping Human in the Loop of Drone-assisted Inspection

Audio commands are a preferred communication medium to keep inspectors in the loop of civil infrastructure inspection performed by a semi-autonomous drone. To understand job-specific commands from a group of heterogeneous and dynamic inspectors, a model needs to be developed cost-effectively for the group and easily adapted when the group changes. This paper is motivated to build a multi-tasking deep learning model that possesses a Share-Split-Collaborate architecture. This architecture allows the two classification tasks to share the feature extractor and then split subject-specific and keyword-specific features intertwined in the extracted features through feature projection and collaborative training. A base model for a group of five authorized subjects is trained and tested on the inspection keyword dataset collected by this study. The model achieved a 95.3% or higher mean accuracy in classifying the keywords of any authorized inspectors. Its mean accuracy in speaker classification is 99.2%. Due to the richer keyword representations that the model learns from the pooled training data, adapting the base model to a new inspector requires only a little training data from that inspector, like five utterances per keyword. Using the speaker classification scores for inspector verification can achieve a success rate of at least 93.9% in verifying authorized inspectors and 76.1\% in detecting unauthorized ones. Further, the paper demonstrates the applicability of the proposed model to larger-size groups on a public dataset. This paper provides a solution to addressing challenges facing AI-assisted human-robot interaction, including worker heterogeneity, worker dynamics, and job heterogeneity.