Hai Liu

Yongjian Deng, Hao Chen, Bochen Xie et al.

Recent advances in event-based research prioritize sparsity and temporal precision. Approaches using dense frame-based representations processed via well-pretrained CNNs are being replaced by the use of sparse point-based representations learned through graph CNNs (GCN). Yet, the efficacy of these graph methods is far behind their frame-based counterparts with two limitations. ($i$) Biased graph construction without carefully integrating variant attributes ($i.e.$, semantics, spatial and temporal cues) for each vertex, leading to imprecise graph representation. ($ii$) Deficient learning because of the lack of well-pretrained models available. Here we solve the first problem by proposing a new event-based GCN (EDGCN), with a dynamic aggregation module to integrate all attributes of vertices adaptively. To address the second problem, we introduce a novel learning framework called cross-representation distillation (CRD), which leverages the dense representation of events as a cross-representation auxiliary to provide additional supervision and prior knowledge for the event graph. This frame-to-graph distillation allows us to benefit from the large-scale priors provided by CNNs while still retaining the advantages of graph-based models. Extensive experiments show our model and learning framework are effective and generalize well across multiple vision tasks.

Hai Liu, Huawei Wu

For an $(n,k,\ell)$ MDS array code over $\mathbb{F}_q$, how small can the repair bandwidth and repair I/O be under linear exact repair? We study this question in the regime where the field size $q$, the redundancy $r=n-k$, and the sub-packetization level $\ell$ are fixed, while the code length $n$ varies, and we develop a geometric approach to this setting. Our starting point is an intrinsic reformulation of linear exact repair for MDS array codes in terms of subspace intersections and, for repair I/O, the projective point configurations induced by a parity-check realization. This viewpoint yields a simple projective counting argument establishing the general lower bound $$β_{\mathrm{avg}},β_{\max},γ_{\mathrm{avg}},γ_{\max}\;\ge\;\ell(n-1)-\frac{q^{(r-1)\ell}-1}{q-1}$$ for linear exact repair of every $(n,k,\ell)$ MDS array code over $\mathbb{F}_q$ with redundancy $r=n-k\ge 2$. To our knowledge, this is the first lower bound of this form that applies to arbitrary redundancy $r\ge 2$ and sub-packetization level $\ell$. At first glance, the projective counting bound appears rather coarse and therefore unlikely to be attained. We prove that this intuition is correct whenever $r\ge 3$ and $\ell\ge 2$. For $r=2$, the picture changes completely. Using Desarguesian spreads from finite geometry, we construct MDS array codes that attain the bound over a broad interval of code lengths, up to the maximum possible length $q^{\ell}+1$, and do so simultaneously for both repair bandwidth and repair I/O. In the smallest nontrivial case $(r,\ell)=(2,2)$, we also prove a converse within the regular-spread model. Together, these results identify a uniform obstruction governing linear exact repair and show that, in the two-parity case, this obstruction is tight.

Yongjian Deng, Hao Chen, Hai Liu et al.

Event cameras attract researchers' attention due to their low power consumption, high dynamic range, and extremely high temporal resolution. Learning models on event-based object classification have recently achieved massive success by accumulating sparse events into dense frames to apply traditional 2D learning methods. Yet, these approaches necessitate heavy-weight models and are with high computational complexity due to the redundant information introduced by the sparse-to-dense conversion, limiting the potential of event cameras on real-life applications. This study aims to address the core problem of balancing accuracy and model complexity for event-based classification models. To this end, we introduce a novel graph representation for event data to exploit their sparsity better and customize a lightweight voxel graph convolutional neural network (\textit{EV-VGCNN}) for event-based classification. Specifically, (1) using voxel-wise vertices rather than previous point-wise inputs to explicitly exploit regional 2D semantics of event streams while keeping the sparsity;(2) proposing a multi-scale feature relational layer (\textit{MFRL}) to extract spatial and motion cues from each vertex discriminatively concerning its distances to neighbors. Comprehensive experiments show that our model can advance state-of-the-art classification accuracy with extremely low model complexity (merely 0.84M parameters).