arXiv subjects

Xiangchen Zhao

Publications and source records attributed to Xiangchen Zhao.

3 recordsLinked to original sources

arXiv2026

From Imitation to Intuition: Intrinsic Reasoning for Open-Instance Video Classification

Conventional video classification models, acting as effective imitators, excel in scenarios with homogeneous data distributions. However, real-world applications often present an open-instance challenge, where intra-class variations are vast and complex, beyond existing benchmarks. While traditional video encoder models struggle to fit these diverse distributions, vision-language models (VLMs) offer superior generalization but have not fully leveraged their reasoning capabilities (intuition) for such tasks. In this paper, we bridge this gap with an intrinsic reasoning framework that evolves open-instance video classification from imitation to intuition. Our approach, namely DeepIntuit, begins with a cold-start supervised alignment to initialize reasoning capability, followed by refinement using Group Relative Policy Optimization (GRPO) to enhance reasoning coherence through reinforcement learning. Crucially, to translate this reasoning into accurate classification, DeepIntuit then introduces an intuitive calibration stage. In this stage, a classifier is trained on this intrinsic reasoning traces generated by the refined VLM, ensuring stable knowledge transfer without distribution mismatch. Extensive experiments demonstrate that for open-instance video classification, DeepIntuit benefits significantly from transcending simple feature imitation and evolving toward intrinsic reasoning. Our project is available at https://bwgzk-keke.github.io/DeepIntuit/.

cs.CV↗

arXiv2022

CenterFormer: Center-based Transformer for 3D Object Detection

Query-based transformer has shown great potential in constructing long-range attention in many image-domain tasks, but has rarely been considered in LiDAR-based 3D object detection due to the overwhelming size of the point cloud data. In this paper, we propose CenterFormer, a center-based transformer network for 3D object detection. CenterFormer first uses a center heatmap to select center candidates on top of a standard voxel-based point cloud encoder. It then uses the feature of the center candidate as the query embedding in the transformer. To further aggregate features from multiple frames, we design an approach to fuse features through cross-attention. Lastly, regression heads are added to predict the bounding box on the output center feature representation. Our design reduces the convergence difficulty and computational complexity of the transformer structure. The results show significant improvements over the strong baseline of anchor-free object detection networks. CenterFormer achieves state-of-the-art performance for a single model on the Waymo Open Dataset, with 73.7% mAPH on the validation set and 75.6% mAPH on the test set, significantly outperforming all previously published CNN and transformer-based methods. Our code is publicly available at https://github.com/TuSimple/centerformer

cs.CV↗

arXiv2021

Design and Experimental Evaluation of Algorithms for Optimizing the Throughput of Dispersed Computing

With growing deployment of Internet of Things (IoT) and machine learning (ML) applications, which need to leverage computation on edge and cloud resources, it is important to develop algorithms and tools to place these distributed computations to optimize their performance. We address the problem of optimally placing computations (described as directed acyclic graphs (DAGs)) on a set of machines to maximize the steady-state throughput for pipelined inputs. Traditionally, such optimization has focused on a different metric, minimizing single-shot makespan, and a well-known algorithm is the Heterogeneous Earliest Finish Time (HEFT) algorithm. Maximizing throughput however, is more suitable for many real-time, edge, cloud and IoT applications, we present a different scheduling algorithm, namely Throughput HEFT (TPHEFT). Further, we present two throughput-oriented enhancements which can be applied to any baseline schedule, that we refer to as "node splitting" (SPLIT) and "task duplication" (DUP). In order to implement and evaluate these algorithms, we built new subsystems and plugins for an open-source dispersed computing framework called Jupiter. Experiments with varying DAG structures indicate that: 1) TPHEFT can significantly improve throughput performance compared to HEFT (up to 2.3 times in our experiments), with greater gains when there is less degree of parallelism in the DAG, 2) Node splitting can potentially improve performance over a baseline schedule, with greater gains when there's an imbalanced allocation of computation or inter-task communication, and 3) Task duplication generally gives improvements only when running upon a baseline that places communication over slow links. To our knowledge, this is the first study to present a systematic experimental implementation and exploration of throughput-enhancing techniques for dispersed computing on real testbeds.

cs.DC↗