Ziwei Wang

Ziwei Wang is a PhD candidate at the Intelligent Vision Group (IVG), Department of Automation, Tsinghua University, advised by Prof. Jiwen Lu. He received the BS degree from the Department of Physics, Tsinghua University, China, in 2018. His research interests include tiny machine learning, robotic vision and scene understanding. He has published over 20 scientific papers in the IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE Robotics and Automation Letters, CVPR, ICCV, ECCV, IROS and ICRA. He serves as a regular reviewer member for the IEEE Transactions on Image Processing, IEEE Transactions on Circuits and Systems for Video Technology, Pattern Recognition Letters, CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, ICRA, WACV, ACCV, ICPR, ICME and ICIP.

Email  /  CV  /  Google Scholar  /  Github

We propose an ultrafast auto- mated model compression framework for flexible network deployment, where we can obtain the optimal compression policy within several seconds.

we propose binary sparse convolutional networks for efficient point cloud analysis, where we search the optimal subset of convolution operation that activates the sparse convolution at various locations for quantization error alleviation.

We propose a category-level shape estimation method for densely cluttered objects, which addresses the challenges of large object segmentation errors and inaccurate shape recovery on unseen instances.

We propose the extremely low-precision vision transformers in 2-4 bits, where the self-attention rank consistency and group-wise quantization are presented for quantization error minimization.

we develop a packing planning method for general objects including the packing sequence, locations and orientations to maximize the space utilization ratio.

we develop a post-hoc visual explanation method based on the Shapley value in class activation mapping.

We propose an interactive exploration framework called Smart Explorer for recognizing all objects in dense clutters.

we present a generic framework for robotic motion planning in dense clutter with diverse action primitives and generator-evaluator architectures.

We propose the unsupervised binary descriptor learning method via dynamic bitwise interaction mining (D-GraphBit), where a graph convolutional network called GraphMiner reasons the optimal bitwise interaction for each input sample.

We propose a Shapley value based method to evaluate operation contribution (Shapley-NAS) for neural architecture search.

We propose a generalizable mixed-precision quantization (GMPQ) method for efficient inference.

We propose an instance similarity learning (ISL) method for unsupervised feature representation.

We propose apresent binary neural networks with automatic information compression (AutoBiDet) to automatically adjust the IB trade-off according to the input complexity.

We present a hierarchical channel-wise interaction based binary convolutional neural networks (HCI-BCNN) method to minimize the quantiztaion error for activations via hierarchical reinforcement learning.

We propose a Deep Hashing method with Active Pairwise Supervision (DH-APS) to learn effective binary codes for image search with limited annotation budget.

We propose an efficient binarized object detector that fully utilizes the representational capacity of the binary neural networks by redundancy removal.

We propose a channel-wise interaction based binary convolutional neural network learning method (CI-BCNN) for efficient inference with minimal information loss.

We present a deep multi-quantization network to learn a data-dependent binarization for unsupervised features.

We propose a GraphBit method to learn deep binary descriptors with enhanced reliability in a directed acyclic graph unsupervisedly.

We present an unsupervised feature learning method called deep binary descriptor with multiquantization (DBD-MQ) for visual matching.

Website Template