Dec 2023 – May 2024 · 5 mos

May 2022 – Dec 2022 · 7 mos

• I worked on few-shot Learning with Large Vision-Language Models. I proposed the category name initialization that can obtain state-of-the-art few-shot image classification performance on 13 benchmarks. This work has been published in ICLRW 2023.

May 2021 – Apr 2022 · 11 mos

• I worked on Google Landmark Image Retrieval, where I fused global and local features with Deep Fusion Transformer. I implemented the whole framework using TensorFlow. This method achieves better accuracy and efficiency by replacing the original two-stage image retrieval procedure: global feature retrieval and local feature reranking.

May 2020 – Dec 2020 · 7 mos · Santa Clara, California, United States

• I worked on learning contrastive representation for semantic correspondence. I propose a multi-level contrastive learning approach for semantic matching, which does not rely on any ImageNet pretrained model. I implemented the whole framework using Pytorch. This proposed method performs favorably against the state-of-the-art approaches. This work has been published in IJCV 2022 and filed US Patent.

May 2019 – Aug 2019 · 3 mos · Mountain View, California, United States

• I worked on optical flow estimation. I proposed a learnable cost volume that can be plugged into any existing flow estimation framework for improving flow estimation accuracy and robustness. This works has been published in ECCV 2020 and filed US patent.

Sep 2018 – Sep 2019 · 1 yr · Cupertino, California, United States · Remote

• I worked on adversarial learning of visual privacy-preserving representation. I developed an adversarial learning framework in Pytorch. This works was published in AAAI 2020.

Aug 2018 – Dec 2023 · 5 yrs 4 mos · Merced, California Area

• I have worked on various research problems on computer vision and machine learning with publications.
• Learning Correspondence from Images and Videos (ECCV 2020, IJCV 2022)
• Model Sparsification with Joint Optimization of Group Convolution and Channel Shuffle (UAI 2021)
• Semi-supervised Learning with Meta-Gradient(AISTATS 2021)
• Neural Architecture Search (IJCV)

Jun 2018 – Aug 2018 · 2 mos · Beijing, China

• I worked on Domain Adaptation in Object Detection.

Jan 2017 – Nov 2017 · 10 mos · Beijing, China

• I have worked on face attribute transfiguration using Generative Adversarial Networks (GANs), published in BMVC 2017 as an oral presentation. I have also worked on quantized neural networks, where I trained a GoogLeNet having 4-bit weights and activations to reach 11.4% in top-5 single-crop error on the ImageNet dataset. This work has been published in ICANN 2017.

Jan 2016 – May 2018 · 2 yrs 4 mos · Beijing, China

• I have worked on several topics with publications.
• Multiple Face Attribute Transfer (ECCV 2018)
• Learning Disentangled Representations (ICLRW 2018)
• Pedestrian Tracking (ICIC 2017)