Learning Box Regression and Mask Segmentation under Long-tailed Distribution with

Gradient Transfusing (CRAT)

Tao Wang, Li Yuan, Jiashi Feng and Xinchao Wang,

IJCV 2024

• We study how box regression and mask segmentation are affected by long-tailed distribution and propose CRAT, which is guided by Fisher to augment tail class training during back-propagation.

Overcoming Classifier Imbalance for Long-tail Object Detection with Balanced Group Softmax

Yu Li, Tao Wang, Bingyi Kang, Sheng Tang, Chunfeng Wang, Jintao Li, Jiashi Feng

CVPR 2020 Oral, [Paper][Code][Video]

• Widely adopted by LVIS challenge 2020 and 2021 top-entries

• We propose a specifically re-designed softmax classification module which further improves over SimCal on long-tail object detection and instance segmentation.

The Devil is in Classification: A Simple Framework for Long-tail Instance Segmentation (SimCal)

Tao Wang, Yu Li, Bingyi Kang, Junnan Li, Junhao Liew, Sheng Tang, Steven Hoi, and Jiashi Feng

ECCV 2020, [Paper][Code][Video]

• Based on our LVIS winner solution, we further extend it and improve the performance by discovering a better initialization strategy.

Joint COCO and Mapillary Workshop at ICCV 2019: LVIS Challenge Track: Classification Calibration for Long-tail Instance Segmentation

Tao Wang, Yu Li, Bingyi Kang, Junnan Li, Junhao Liew, Sheng Tang, Steven Hoi, Jiashi Feng

Winner solution for the 1st LVIS challenge at ICCV 2019 [Tech Report]

• Zero-Shot Aerial Object Detection with Visual Description Regularization (DescReg)

• 
  

• Zhengqing Zang, Chenyu Lin, Chenwei Tang, Tao Wang†(Corresponding Author), Jiancheng Lv

• 
  

• AAAI 2024, [Paper][Code]

• We identify the weak semantic-visual correlation challenge in zero-shot aerial object detection domain and propose a visual description regularization method to improve zero-shot detection.

Learning to Detect and Segment for Open Vocabulary Object Detection (CondHead) 

Tao Wang, Nan Li

CVPR 2023, [Paper][Code]

• CondHead conditions the bounding box regression and mask segmentation on the text embeddings, to facilitate open vocabulary object detection.

PoseTriplet: Co-evolving 3D Human Pose Estimation, Imitation, and Hallucination under Self-supervision

Kehong Gong*, Bingbing Li*, Jianfeng Zhang*, Tao Wang*, Jing Huang, Michael Bi Mi, Jiashi Feng, Xinchao Wang (* equal contribution)

CVPR 2022 Oral, [Paper][Code][Video]

• We construct an effective self-supervised framework for 3D human pose estimation, it is self-improving by generating physically plausible 2D-3D training pose data.

Revisiting Knowledge Distillation via Label Smoothing Regularization

Li Yuan, Francis EH Tay, Guilin Li, Tao Wang, Jiashi Feng

CVPR 2020 Oral, [Paper][Code][Video]

• We reveal that knowledge distillation (KD) works as a learned label smoothing regularization, and further propose a novel Teacher-free Knowledge Distillation (Tf-KD) framework.

Distilling Object Detectors with Fine-grained Feature Imitation

Tao Wang, Li Yuan, Xiaopeng Zhang, Jiashi Feng

CVPR 2019, [Paper][Code]

• Highly cited work for knowledge distillation of object detection model.

• We develop a knowledge distillation (KD) framework for object detection, based on feature-level imitation of the estimated foreground object regions.

Few-shot Adaptive Faster R-CNN

Tao Wang, Xiaopeng Zhang, Li Yuan, Jiashi Feng

CVPR 2019, [Paper][Code]

• We reveal that knowledge distillation (KD) works as a learned label smoothing regularization, and further propose a novel Teacher-free Knowledge Distillation (Tf-KD) framework.

SODAR: Segmenting Objects by Dynamically Aggregating Neighboring Mask Representations

Tao Wang, Jun Hao Liew, Yu Li, Yunpeng Chen, Jiashi Feng

TIP 2022, [Paper][Code]

• We reveal the usefulness of neighboring mask predictions and introduce a simple and efficient neighbor aggregation method to improve dense instance segmentation models.

Detect Multi-person with 3D Pose Directly from Multi-view images (Multi-view Pose Transformer, MvP)

Tao Wang, Jianfeng Zhang, Yujun Cai, Shuicheng Yan, Jiashi Feng

NeurIPS 2021, [Paper][Code][Industrial Recognition by XRMoCap][Video][Slides]

• We develop a simple transformer algorithm that directly detects multi-person and predicts their 3D pose from multi-view images.

Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet

Li Yuan*, Yunpeng Chen, Tao Wang*, Weihao Yu, Yujun Shi, Zihang Jiang, Francis E.H. Tay, Jiashi Feng, Shuicheng Yan (Work done during internship at Yitu)

ICCV 2021, [Paper][Code][Video][Most Influential ICCV Papers]

• We introduce a Tokens-to-Token (T2T) transformation scheme to progressively structurize the image to tokens by recursively aggregating neighboring Tokens.

PnP-DETR: Towards Efficient Visual Analysis with Transformers

Tao Wang, Li Yuan, Yunpeng Chen, Jiashi Feng, Shuicheng Yan

ICCV 2021, [Paper][Code][Video]

• We propose Poll and Pool sampling to reduce the spatial redundancy of image features for efficient transformer processing.