Domain Adaptation in Computer Vision

In recent years, while we are trying to attain the longterm computer vision task of creating smart machines that learn by visual exploration of world. We are currently trying to build on the immediate goal of learning classifiers and recognizing objects in the real world. A common problem that we face is that we require large amount of labeled data for this and at times the tasks we may work on may change.

The paradigm of domain adaptation has burgeoned for solving this challenge using techniques that try generalizing the knowledge of previous tasks and models in Computer Vision has also been seen. The goal is to understand this application of Domain Adaptation through research publications in leading conferences in the field of Computer Vision.

Some of the tasks we will be going through would be ranging from to computer vision, ranging from image classification, semantic segmentation, object detection, face recognition, person re-identification, video understanding and 3D reconstruction.

List of Papers

1. Divergence-based Domain Adaptation

[a] MMD - Maximum Mean Discrepancy

• Learning Transferable Features with Deep Adaptation Networks [Paper] [Review]
  • Mingsheng Long, Yue Cao, Jianmin Wang, Michael I. Jordan. ICML'15
• Deep Transfer Learning with Joint Adaptation Networks [Paper] [Review]
  • Mingsheng Long, Han Zhu, Jianmin Wang, Michael I. Jordan. ICML'17
• Unsupervised Domain Adaptation with Imbalanced Cross-Domain Data [Paper] [Review]
  • Tzu-Ming Harry Hsu, Wei-Yu Chen, Cheng-An Hou, Yao-Hung Hubert Tsai, Yi-Ren Yeh, and Yu-Chiang Frank Wang, ICCV'15

[b] CORAL - Correlation Alignment

• Minimal-Entropy Correlation Alignment for Unsupervised Deep Domain Adaptation [Paper] [Review]
  • Pietro Morerio, Jacopo Cavazza, Vittorio Murino. ICLR'18

[c] CCD - Contrastive Domain Discrepancy

[d] Wasserstein metric

• Joint Distribution Optimal Transportation for Domain Adaptation [Paper] [Review]
  • Nicolas Courty, Rémi Flamary, Amaury Habrard, Alain Rakotomamonjy, NeurIPS‘17

2. Adversarial-based Domain Adaptation

[e] Generative

• Unsupervised Cross-Domain Image Generation [Paper] [Review]
  • Yaniv Taigman, Adam Polyak, Lior Wolf, ICLR‘17
• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks [Paper] [Review]
  • Jun-Yan Zhu and Taesung Park, Phillip Isola and Alexei A. Efros, ICCV‘17
• InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets [Paper] [Review]
  • Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, Pieter Abbeel, NeurIPS'16

[f] Non-generative

• Domain Adaptive Faster R-CNN for Object Detection in the Wild[Paper] [Review]

  • Yuhua Chen, Wen Li, Christos Sakaridis, Dengxin Dai, Luc Van Gool, CVPR‘18

• Label Efficient Learning of Transferable Representations across Domains and Tasks[Paper] [Review]

  • Zelun Luo, Yuliang Zou, Judy Hoffman, Li Fei-Fei, NeurIPS'17

• DANN

• ADDA

3. Reconstruction-based Domain Adaptation

[g] Encoder-Decoder

• Domain Separation Networks [Paper] [Review]
  • Konstantinos Bousmalis, George Trigeorgis, Nathan Silberman, Dilip Krishnan, Dumitru Erhan, NeurIPS'16
• DRCN - Reconstruction Classification

[h] Generative based approaches

Resources

1. Tutorial on Transfer learning & Domain adaptation