Publications

  • Visual Prompt Tuning
    Menglin Jia*, Luming Tang*, Bor-Chun Chen, Claire Cardie, Serge Belongie
    In ECCV, 2022
    TLDR : Learning a prompt is the best way to transfer vision transformers to new tasks
    pdf    bibtex
  • Exploiting Playbacks in Unsupervised Domain Adaptation for 3D Object Detection
    Yurong You*, Carlos Andres Diaz-Ruiz*, Yan Wang, Wei-Lun Chao, Bharath Hariharan, Mark Campbell, Kilian Weinberger
    In ICRA, 2022
    TLDR : Extrapolating tracks of detected objects yields good ground truth for adapting 3D detectors
    pdf    bibtex
  • Learning to Detect Mobile Objects from LiDAR Scans Without Labels
    Yurong You*, Katie Luo*, Cheng Perng Phoo, Wei-Lun Chao, Wen Sun, Bharath Hariharan, Mark Campbell, Kilian Weinberger
    In CVPR, 2022
    TLDR : Can we train a 3D object detector without labels by simply driving around?
    pdf    bibtex
  • Coarsely-labeled Data for Better Few-shot Transfer
    Cheng Perng Phoo, Bharath Hariharan
    In ICCV, 2021
    TLDR : Coarse labels are cheap to acquire and can boost few-shot learning
    pdf    bibtex
  • Field Guide-inspired Zero-Shot Learning
    Utkarsh Mall, Bharath Hariharan, Kavita Bala
    In ICCV, 2021
    TLDR : A more usable, active-learning based interface for specifying hundreds of attributes in zero-shot learning
    pdf    bibtex
  • Can We Characterize Tasks Without Labels or Features?
    Bram Wallace, Ziyang Wu, Bharath Hariharan
    In CVPR, 2021
    TLDR : You can pick the best pre-training task for novel tasks without any labels, and in new domains without any pre-trained networks.
    pdf    bibtex
  • Few-Shot Classification with Feature Map Reconstruction Networks
    Davis Wertheimer, Luming Tang, Bharath Hariharan
    In CVPR, 2021
    TLDR : We assign test images to the training class that yields the best reconstruction of its feature map. This builds in pose invariance, preserves details and leads to state-of-the-art few-shot learning.
    pdf    project page    bibtex
  • PiCIE: Unsupervised Semantic Segmentation using Invariance and Equivariance in Clustering
    Jang Hyun Cho, Utkarsh Mall, Kavita Bala, Bharath Hariharan
    In CVPR, 2021
    TLDR : We can train semantic segmentation models without pixel labels by clustering while encouraging equivariance to geometric transforms and invariance to photometric ones.
    project page    bibtex
  • Self-training For Few-shot Transfer Across Extreme Task Differences
    Cheng Perng Phoo, Bharath Hariharan
    In ICLR, 2021 (Oral)
    TLDR : We can build state-of-the-art neural feature representations for new domains by (self)training students to replicate pseudo-labels produced by a teacher from another, unrelated problem domain.
    pdf    bibtex
  • When Does Self-supervision Improve Few-shot Learning?
    Jong-Chyi Su, Subhransu Maji, Bharath Hariharan
    In ECCV, 2020
    TLDR : An analysis of if and when auxilliary losses based on self-supervision aid representation learning for few-shot learning
    pdf    project page    bibtex
  • Extending and Analyzing Self-Supervised Learning Across Domains
    Bram Wallace, Bharath Hariharan
    In ECCV, 2020
    TLDR : An analysis of how well self-supervision techniques generalize across natural and non-natural domains
    pdf    bibtex
  • Learning Feature Descriptors using Camera Pose Supervision
    Qianqian Wang, Xiaowei Zhou, Bharath Hariharan, Noah Snavely
    In ECCV, 2020 (Oral)
    TLDR : We learn neural network-based feature descriptors solely from camera pose without any ground truth correspondences
    pdf    supp    project page    bibtex
  • Revisiting Pose-Normalization for Fine-Grained Few-Shot Recognition
    Luming Tang, Davis Wertheimer, Bharath Hariharan
    In CVPR, 2020
    TLDR : Pose normalization by explicitly featurizing the appearance of object parts substantially improves few-shot learning for fine-grained classification problems
    pdf    supp    code    bibtex
  • Train in Germany, Test in The USA: Making 3D Object Detectors Generalize
    Yan Wang, Xiangyu Chen, Yurong You, Li Erran Li, Bharath Hariharan, Mark Campbell, Kilian Q. Weinberger, Wei-Lun Chao
    In CVPR, 2020
    TLDR : Domain differences lead to catastrophic failures in 3D object detection. We present a simple and effective remedy.
    pdf    supp    bibtex
  • Few-Shot Generalization for Single-Image 3D Reconstruction via Prior
    Bram Wallace, Bharath Hariharan
    In ICCV, 2019
    TLDR : A single-view 3D reconstruction system that explicitly leverages the average category shape, and can thus generalize to new classes without training
    pdf    bibtex
  • Few-shot Learning with Localization in Realistic Settings
    Davis Wertheimer, Bharath Hariharan
    In CVPR, 2019 (Oral)
    TLDR : We take few-shot learning beyond constrained benchmarks to real-world problems with class imbalance, fine-grained distinctions and potentially additional annotations
    pdf    supp    code    bibtex
  • Learning Single-View 3D Reconstruction with Limited Pose Supervision
    Guandao Yang, Yin Cui, Serge Belongie, Bharath Hariharan
    In ECCV, 2018
    TLDR : We explore many ways of reducing the need for ground truth annotations for training 3D reconstruction systems
    pdf    bibtex
  • Low-shot Learning from Imaginary Data
    Yu-Xiong Wang, Ross Girshick, Martial Herbert, Bharath Hariharan
    In CVPR, 2018 (Spotlight)
    TLDR : We train few-shot learners in an end-to-end manner to imagine additional training data.
    pdf    bibtex
  • Low-shot learning with large-scale diffusion
    Matthijs Douze, Arthur Szlam, Bharath Hariharan, Hervé Jégou
    In CVPR, 2018
    TLDR : One of the first explorations into transductive few-shot learning
    pdf    bibtex
  • Low-shot Visual Recognition by Shrinking and Hallucinating Features
    Bharath Hariharan, Ross Girshick
    In ICCV, 2017 (Spotlight)
    TLDR : We create the first large-scale few-shot learning benchmark and propose two baseline techniques
    pdf    supp    code    bibtex
  • Learning Features by Watching Objects Move
    Deepak Pathak, Ross Girshick, Piotr Dollár, Trevor Darrell, Bharath Hariharan
    In CVPR, 2017
    TLDR : A self-supervised learning method that trains neural networks to replicate the output of motion segmentation, but from a single frame, forcing the network to recognize objects that tend to move
    pdf    project page    bibtex