Searcharxiv⌕ Search

arXiv subjects

Hong Cai

Publications and source records attributed to Hong Cai.

67 records · Page 4Linked to original sources
arXiv2021

HS3: Learning with Proper Task Complexity in Hierarchically Supervised Semantic Segmentation

While deeply supervised networks are common in recent literature, they typically impose the same learning objective on all transitional layers despite their varying representation powers. In this paper, we propose Hierarchically Supervised Semantic Segmentation (HS3), a training scheme that supervises intermediate layers in a segmentation network to learn meaningful representations by varying task complexity. To enforce a consistent performance vs. complexity trade-off throughout the network, we derive various sets of class clusters to supervise each transitional layer of the network. Furthermore, we devise a fusion framework, HS3-Fuse, to aggregate the hierarchical features generated by these layers, which can provide rich semantic contexts and further enhance the final segmentation. Extensive experiments show that our proposed HS3 scheme considerably outperforms vanilla deep supervision with no added inference cost. Our proposed HS3-Fuse framework further improves segmentation predictions and achieves state-of-the-art results on two large segmentation benchmarks: NYUD-v2 and Cityscapes.

cs.CV↗
arXiv2021

AuxAdapt: Stable and Efficient Test-Time Adaptation for Temporally Consistent Video Semantic Segmentation

In video segmentation, generating temporally consistent results across frames is as important as achieving frame-wise accuracy. Existing methods rely either on optical flow regularization or fine-tuning with test data to attain temporal consistency. However, optical flow is not always avail-able and reliable. Besides, it is expensive to compute. Fine-tuning the original model in test time is cost sensitive. This paper presents an efficient, intuitive, and unsupervised online adaptation method, AuxAdapt, for improving the temporal consistency of most neural network models. It does not require optical flow and only takes one pass of the video. Since inconsistency mainly arises from the model's uncertainty in its output, we propose an adaptation scheme where the model learns from its own segmentation decisions as it streams a video, which allows producing more confident and temporally consistent labeling for similarly-looking pixels across frames. For stability and efficiency, we leverage a small auxiliary segmentation network (AuxNet) to assist with this adaptation. More specifically, AuxNet readjusts the decision of the original segmentation network (Main-Net) by adding its own estimations to that of MainNet. At every frame, only AuxNet is updated via back-propagation while keeping MainNet fixed. We extensively evaluate our test-time adaptation approach on standard video benchmarks, including Cityscapes, CamVid, and KITTI. The results demonstrate that our approach provides label-wise accurate, temporally consistent, and computationally efficient adaptation (5+ folds overhead reduction comparing to state-of-the-art test-time adaptation methods).

cs.CV↗
arXiv2021

Perceptual Consistency in Video Segmentation

In this paper, we present a novel perceptual consistency perspective on video semantic segmentation, which can capture both temporal consistency and pixel-wise correctness. Given two nearby video frames, perceptual consistency measures how much the segmentation decisions agree with the pixel correspondences obtained via matching general perceptual features. More specifically, for each pixel in one frame, we find the most perceptually correlated pixel in the other frame. Our intuition is that such a pair of pixels are highly likely to belong to the same class. Next, we assess how much the segmentation agrees with such perceptual correspondences, based on which we derive the perceptual consistency of the segmentation maps across these two frames. Utilizing perceptual consistency, we can evaluate the temporal consistency of video segmentation by measuring the perceptual consistency over consecutive pairs of segmentation maps in a video. Furthermore, given a sparsely labeled test video, perceptual consistency can be utilized to aid with predicting the pixel-wise correctness of the segmentation on an unlabeled frame. More specifically, by measuring the perceptual consistency between the predicted segmentation and the available ground truth on a nearby frame and combining it with the segmentation confidence, we can accurately assess the classification correctness on each pixel. Our experiments show that the proposed perceptual consistency can more accurately evaluate the temporal consistency of video segmentation as compared to flow-based measures. Furthermore, it can help more confidently predict segmentation accuracy on unlabeled test frames, as compared to using classification confidence alone. Finally, our proposed measure can be used as a regularizer during the training of segmentation models, which leads to more temporally consistent video segmentation while maintaining accuracy.

cs.CV↗
arXiv2021

X-Distill: Improving Self-Supervised Monocular Depth via Cross-Task Distillation

In this paper, we propose a novel method, X-Distill, to improve the self-supervised training of monocular depth via cross-task knowledge distillation from semantic segmentation to depth estimation. More specifically, during training, we utilize a pretrained semantic segmentation teacher network and transfer its semantic knowledge to the depth network. In order to enable such knowledge distillation across two different visual tasks, we introduce a small, trainable network that translates the predicted depth map to a semantic segmentation map, which can then be supervised by the teacher network. In this way, this small network enables the backpropagation from the semantic segmentation teacher's supervision to the depth network during training. In addition, since the commonly used object classes in semantic segmentation are not directly transferable to depth, we study the visual and geometric characteristics of the objects and design a new way of grouping them that can be shared by both tasks. It is noteworthy that our approach only modifies the training process and does not incur additional computation during inference. We extensively evaluate the efficacy of our proposed approach on the standard KITTI benchmark and compare it with the latest state of the art. We further test the generalizability of our approach on Make3D. Overall, the results show that our approach significantly improves the depth estimation accuracy and outperforms the state of the art.

cs.CV↗
arXiv2020

Uniqueness of Dissipative Solution for Camassa-Holm Equation with Peakon-Antipeakon Initial Data

We give a proof for the uniqueness of dissipative solution for the Camassa-Holm equation with some peakon-antipeakon initial data following Dafermos' earlier resut in [5] on the Hunter-Saxton equation. Our result shows that two existing global existence frameworks, through the vanishing viscosity method by Xin-Zhang in [11] and the transformation of coordinate method for dissipative solutions by Bressan-Constantin in [3], give the same solution, for a special but typical initial data forming finite time gradient blowup.

math.AP↗
arXiv2020

A Finsler type Lipschitz optimal transport metric for a quasilinear wave equation

We consider the global well-posedness of weak energy conservative solution to a general quasilinear wave equation through variational principle, where the solution may form finite time cusp singularity, when energy concentrates. As a main result in this paper, we construct a Finsler type optimal transport metric, then prove that the solution flow is Lipschitz under this metric. We also prove a generic regularity result by applying Thom's transversality theorem, then find piecewise smooth transportation paths among a dense set of solutions. The results in this paper are for large data solutions, without restriction on the size of solutions.

math.AP↗
arXiv2020

Singularity formation for radially symmetric expanding wave of Compressible Euler Equations

In this paper, for compressible Euler equations in multiple space dimensions, we prove the break-down of classical solutions with a large class of initial data by tracking the propagation of radially symmetric expanding wave including compression. The singularity formation is corresponding to the finite time shock formation. We also provide some new global sup-norm estimates on velocity and density functions for classical solutions. The results in this paper have no restriction on the size of solutions, hence are large data results.

math.AP↗
arXiv2019

Fisher-KPP dynamics in diffusive Rosenzweig-MacArthur and Holling-Tanner models

We prove the existence of traveling fronts in diffusive Rosenzweig-MacArthur and Holling-Tanner population models and investigate their relation with fronts in a scalar Fisher-KPP equation. More precisely, we prove the existence of fronts in a Rosenzweig-MacArthur predator-prey model in two situations: when the prey diffuses at the rate much smaller than that of the predator and when both the predator and the prey diffuse very slowly. Both situations are captured as singular perturbations of the associated limiting systems. In the first situation we demonstrate clear relations of the fronts with the fronts in a scalar Fisher-KPP equation. Indeed, we show that the underlying dynamical system in a singular limit is reduced to a scalar Fisher-KPP equation and the fronts supported by the full system are small perturbations of the Fisher-KPP fronts. We obtain a similar result for a diffusive Holling-Tanner population model. In the second situation for the Rosenzweig-MacArthur model we prove the existence of the fronts but without observing a direct relation with Fisher-KPP equation. The analysis suggests that, in a variety of reaction-diffusion systems that rise in population modeling, parameter regimes may be found when the dynamics of the system is inherited from the scalar Fisher-KPP equation.

math.AP↗
arXiv2018

PieAPP: Perceptual Image-Error Assessment through Pairwise Preference

The ability to estimate the perceptual error between images is an important problem in computer vision with many applications. Although it has been studied extensively, however, no method currently exists that can robustly predict visual differences like humans. Some previous approaches used hand-coded models, but they fail to model the complexity of the human visual system. Others used machine learning to train models on human-labeled datasets, but creating large, high-quality datasets is difficult because people are unable to assign consistent error labels to distorted images. In this paper, we present a new learning-based method that is the first to predict perceptual image error like human observers. Since it is much easier for people to compare two given images and identify the one more similar to a reference than to assign quality scores to each, we propose a new, large-scale dataset labeled with the probability that humans will prefer one image over another. We then train a deep-learning model using a novel, pairwise-learning framework to predict the preference of one distorted image over the other. Our key observation is that our trained network can then be used separately with only one distorted image and a reference to predict its perceptual error, without ever being trained on explicit human perceptual-error labels. The perceptual error estimated by our new metric, PieAPP, is well-correlated with human opinion. Furthermore, it significantly outperforms existing algorithms, beating the state-of-the-art by almost 3x on our test set in terms of binary error rate, while also generalizing to new kinds of distortions, unlike previous learning-based methods.

cs.CV↗
arXiv2017

Metropolitan quantum key distribution with silicon photonics

Photonic integrated circuits (PICs) provide a compact and stable platform for quantum photonics. Here we demonstrate a silicon photonics quantum key distribution (QKD) transmitter in the first high-speed polarization-based QKD field tests. The systems reach composable secret key rates of 950 kbps in a local test (on a 103.6-m fiber with a total emulated loss of 9.2 dB) and 106 kbps in an intercity metropolitan test (on a 43-km fiber with 16.4 dB loss). Our results represent the highest secret key generation rate for polarization-based QKD experiments at a standard telecom wavelength and demonstrate PICs as a promising, scalable resource for future formation of metropolitan quantum-secure communications networks.

quant-ph↗
arXiv2016

Lipschitz metric for the Novikov equation

We consider the Lipschitz continuous dependence of solutions for the Novikov equation with respect to the initial data. In particular, we construct a Finsler type optimal transport metric which renders the solution map Lipschitz continuous on bounded set of $H^1(R)\cap W^{1,4}(R)$, although it is not Lipschitz continuous under the natural Sobolev metric from energy law due to the finite time gradient blowup. By an application of Thom's transversality Theorem, we also prove that when the initial data are in an open dense set of $H^1(R)\cap W^{1,4}(R)$, the solution is piecewise smooth. This generic regularity result helps us extend the Lipschitz continuous metric to the general weak solutions.

math.AP↗
arXiv2016

Motion and Communication Co-optimization with Path Planning and Online Channel Estimation

This paper considers the problem of optimally balancing motion energy and communication transmission energy of a mobile robot tasked with transmitting a given number of data bits to a remote station, while navigating to a prespecified destination in a given amount of time. The problem is cast in the setting of optimal control, where the robot has to choose its path, acceleration, and transmission rate along the path so as to minimize its energy required for transmission and motion, while satisfying various power and communication constraints. We use realistic models for the robot's channel estimation, motion dynamics, and power and energy costs. The main contribution of the paper is to show how to co-optimize robot's path along with other communication and motion variables. Two versions of the problem are solved: the first is defined offline by assuming that all the channel measurements are taken before the robots starts moving, while in the second the channel estimation is updated while the robot is in motion, and hence it is solved online. In both cases we utilize an in-house algorithm that computes near-optimal solutions in little time, which enables its use in the online setting. The optimization strategy is described in detail and validated by simulation of realistic scenarios.

math.OC↗