Suhas Lohit

Suhas Lohit
  • Biography

    Before coming to MERL, Suhas worked as an intern at MERL (summer 2018), SRI International (summer 2017) and Nvidia (summer 2016). His research interests include computer vision, computational imaging and deep learning. Recently, his research focus has been on creating hybrid model- and data-driven neural architectures for various applications in imaging and vision. He won the Best Paper Award at the CVPR workshop on Computational Cameras and Displays in 2015 and the University Graduate Fellowship at ASU for 2015-16.

  • Recent News & Events

    •  NEWS    MERL researchers presenting five papers at NeurIPS 2022
      Date: November 29, 2022 - December 9, 2022
      Where: NeurIPS 2022
      MERL Contacts: Moitreya Chatterjee; Anoop Cherian; Michael J. Jones; Suhas Lohit
      Research Areas: Artificial Intelligence, Computer Vision, Machine Learning, Speech & Audio
      Brief
      • MERL researchers are presenting 5 papers at the NeurIPS Conference, which will be held in New Orleans from Nov 29-Dec 1st, with virtual presentations in the following week. NeurIPS is one of the most prestigious and competitive international conferences in machine learning.

        MERL papers in NeurIPS 2022:

        1. “AVLEN: Audio-Visual-Language Embodied Navigation in 3D Environments” by Sudipta Paul, Amit Roy-Chowdhary, and Anoop Cherian

        This work proposes a unified multimodal task for audio-visual embodied navigation where the navigating agent can also interact and seek help from a human/oracle in natural language when it is uncertain of its navigation actions. We propose a multimodal deep hierarchical reinforcement learning framework for solving this challenging task that allows the agent to learn when to seek help and how to use the language instructions. AVLEN agents can interact anywhere in the 3D navigation space and demonstrate state-of-the-art performances when the audio-goal is sporadic or when distractor sounds are present.

        2. “Learning Partial Equivariances From Data” by David W. Romero and Suhas Lohit

        Group equivariance serves as a good prior improving data efficiency and generalization for deep neural networks, especially in settings with data or memory constraints. However, if the symmetry groups are misspecified, equivariance can be overly restrictive and lead to bad performance. This paper shows how to build partial group convolutional neural networks that learn to adapt the equivariance levels at each layer that are suitable for the task at hand directly from data. This improves performance while retaining equivariance properties approximately.

        3. “Learning Audio-Visual Dynamics Using Scene Graphs for Audio Source Separation” by Moitreya Chatterjee, Narendra Ahuja, and Anoop Cherian

        There often exist strong correlations between the 3D motion dynamics of a sounding source and its sound being heard, especially when the source is moving towards or away from the microphone. In this paper, we propose an audio-visual scene-graph that learns and leverages such correlations for improved visually-guided audio separation from an audio mixture, while also allowing predicting the direction of motion of the sound source.

        4. “What Makes a "Good" Data Augmentation in Knowledge Distillation - A Statistical Perspective” by Huan Wang, Suhas Lohit, Michael Jones, and Yun Fu

        This paper presents theoretical and practical results for understanding what makes a particular data augmentation technique (DA) suitable for knowledge distillation (KD). We design a simple metric that works very well in practice to predict the effectiveness of DA for KD. Based on this metric, we also propose a new data augmentation technique that outperforms other methods for knowledge distillation in image recognition networks.

        5. “FeLMi : Few shot Learning with hard Mixup” by Aniket Roy, Anshul Shah, Ketul Shah, Prithviraj Dhar, Anoop Cherian, and Rama Chellappa

        Learning from only a few examples is a fundamental challenge in machine learning. Recent approaches show benefits by learning a feature extractor on the abundant and labeled base examples and transferring these to the fewer novel examples. However, the latter stage is often prone to overfitting due to the small size of few-shot datasets. In this paper, we propose a novel uncertainty-based criteria to synthetically produce “hard” and useful data by mixing up real data samples. Our approach leads to state-of-the-art results on various computer vision few-shot benchmarks.
    •  
  • Awards

    •  AWARD    Best Paper - Honorable Mention Award at WACV 2021
      Date: January 6, 2021
      Awarded to: Rushil Anirudh, Suhas Lohit, Pavan Turaga
      MERL Contact: Suhas Lohit
      Research Areas: Computational Sensing, Computer Vision, Machine Learning
      Brief
      • A team of researchers from Mitsubishi Electric Research Laboratories (MERL), Lawrence Livermore National Laboratory (LLNL) and Arizona State University (ASU) received the Best Paper Honorable Mention Award at WACV 2021 for their paper "Generative Patch Priors for Practical Compressive Image Recovery".

        The paper proposes a novel model of natural images as a composition of small patches which are obtained from a deep generative network. This is unlike prior approaches where the networks attempt to model image-level distributions and are unable to generalize outside training distributions. The key idea in this paper is that learning patch-level statistics is far easier. As the authors demonstrate, this model can then be used to efficiently solve challenging inverse problems in imaging such as compressive image recovery and inpainting even from very few measurements for diverse natural scenes.
    •  
    See All Awards for MERL
  • Internships with Suhas

    • CV1925: Symmetries, equivariance and invariance in deep learning

      MERL is seeking an intern to conduct research in the areas of learning symmetries from data and equivariant neural networks for applications in computer vision. The ideal candidate is a PhD student with experience in deep learning and computer vision and a good publication record at top-tier venues. Prior knowledge and experience with group theory/geometry and equivariant neural networks are a big plus. Good Python and Pytorch/Tensorflow skills are required. Publication of results in conference proceedings and journals is expected. The expected duration of the internship is 3 months and the start date is flexible.

    • CV1927: Deep Multimodal Learning for 3D Vision

      MERL is seeking an intern to conduct research in the area of multimodal 3D vision using modalities such as RGB images and LIDAR. The focus will be on building novel learning algorithms for core applications like robust object detection and semantic segmentation. A good candidate is a PhD student with experience in deep learning and computer vision with a publication record. Prior knowledge and experience in one or more of the above areas are strongly preferred. Good Python and Pytorch/Tensorflow skills are required. Publication of results in conference proceedings and journals is expected. The expected duration of the internship is 3 months and the start date is flexible.

    See All Internships at MERL
  • MERL Publications

    •  Wang, H., Lohit, S., Jones, M.J., Fu, R., "What Makes a “Good” Data Augmentation in Knowledge Distillation – A Statistical Perspective", Advances in Neural Information Processing Systems (NeurIPS), November 2022.
      BibTeX TR2022-147 PDF
      • @inproceedings{Wang2022nov,
      • author = {Wang, Huan and Lohit, Suhas and Jones, Michael J. and Fu, Raymond},
      • title = {What Makes a “Good” Data Augmentation in Knowledge Distillation – A Statistical Perspective},
      • booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
      • year = 2022,
      • month = nov,
      • url = {https://www.merl.com/publications/TR2022-147}
      • }
    •  Romero, D., Lohit, S., "Learning Partial Equivariances from Data", Advances in Neural Information Processing Systems (NeurIPS), November 2022.
      BibTeX TR2022-148 PDF Presentation
      • @inproceedings{Romero2022nov,
      • author = {Romero, David and Lohit, Suhas},
      • title = {Learning Partial Equivariances from Data},
      • booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
      • year = 2022,
      • month = nov,
      • url = {https://www.merl.com/publications/TR2022-148}
      • }
    •  Ahmed, S.M., Lohit, S., Peng, K.-C., Jones, M.J., Roy Chowdhury, A.K., "Cross-Modal Knowledge Transfer Without Task-Relevant Source Data", European Conference on Computer Vision (ECCV), October 2022.
      BibTeX TR2022-135 PDF Video Software Presentation
      • @inproceedings{Ahmed2022oct,
      • author = {Ahmed, Sk Miraj and Lohit, Suhas and Peng, Kuan-Chuan and Jones, Michael J. and Roy Chowdhury, Amit K},
      • title = {Cross-Modal Knowledge Transfer Without Task-Relevant Source Data},
      • booktitle = {European Conference on Computer Vision (ECCV)},
      • year = 2022,
      • month = oct,
      • url = {https://www.merl.com/publications/TR2022-135}
      • }
    •  Mansour, H., Lohit, S., Boufounos, P.T., "Distributed Radar Autofocus Imaging Using Deep Priors", IEEE International Conference on Image Processing (ICIP), October 2022.
      BibTeX TR2022-129 PDF Video
      • @inproceedings{Mansour2022oct,
      • author = {Mansour, Hassan and Lohit, Suhas and Boufounos, Petros T.},
      • title = {Distributed Radar Autofocus Imaging Using Deep Priors},
      • booktitle = {IEEE International Conference on Image Processing (ICIP)},
      • year = 2022,
      • month = oct,
      • url = {https://www.merl.com/publications/TR2022-129}
      • }
    •  Lohit, S., Jones, M.J., "Model Compression Using Optimal Transport", IEEE Winter Conference on Applications of Computer Vision (WACV), January 2022.
      BibTeX TR2022-006 PDF Presentation
      • @inproceedings{Lohit2022jan,
      • author = {Lohit, Suhas and Jones, Michael J.},
      • title = {Model Compression Using Optimal Transport},
      • booktitle = {IEEE Winter Conference on Applications of Computer Vision (WACV)},
      • year = 2022,
      • month = jan,
      • publisher = {CVF OpenAccess},
      • url = {https://www.merl.com/publications/TR2022-006}
      • }
    See All Publications for Suhas
  • Other Publications

    •  Suhas Lohit, Qiao Wang and Pavan Turaga, "Temporal Transformer Networks: Joint Learning of Invariant and Discriminative Time Warping", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019, pp. 12426-12435.
      BibTeX
      • @Inproceedings{lohit2019temporal,
      • author = {Lohit, Suhas and Wang, Qiao and Turaga, Pavan},
      • title = {Temporal Transformer Networks: Joint Learning of Invariant and Discriminative Time Warping},
      • booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
      • year = 2019,
      • pages = {12426--12435}
      • }
    •  Suhas Lohit, Kuldeep Kulkarni, Ronan Kerviche, Pavan Turaga and Amit Ashok, "Convolutional neural networks for noniterative reconstruction of compressively sensed images", IEEE Transactions on Computational Imaging, Vol. 4, No. 3, pp. 326-340, 2018.
      BibTeX
      • @Article{lohit2018convolutional,
      • author = {Lohit, Suhas and Kulkarni, Kuldeep and Kerviche, Ronan and Turaga, Pavan and Ashok, Amit},
      • title = {Convolutional neural networks for noniterative reconstruction of compressively sensed images},
      • journal = {IEEE Transactions on Computational Imaging},
      • year = 2018,
      • volume = 4,
      • number = 3,
      • pages = {326--340},
      • publisher = {IEEE}
      • }
    •  Suhas Lohit, Ankan Bansal, Nitesh Shroff, Jaishanker Pillai, Pavan Turaga and Rama Chellappa, "Predicting Dynamical Evolution of Human Activities from a Single Image", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2018, pp. 383-392.
      BibTeX
      • @Inproceedings{lohit2018predicting,
      • author = {Lohit, Suhas and Bansal, Ankan and Shroff, Nitesh and Pillai, Jaishanker and Turaga, Pavan and Chellappa, Rama},
      • title = {Predicting Dynamical Evolution of Human Activities from a Single Image},
      • booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
      • year = 2018,
      • pages = {383--392}
      • }
    •  Suhas Lohit and Pavan Turaga, "Learning invariant Riemannian geometric representations using deep nets", Proceedings of the IEEE International Conference on Computer Vision Workshops, 2017, pp. 1329-1338.
      BibTeX
      • @Inproceedings{lohit2017learning,
      • author = {Lohit, Suhas and Turaga, Pavan},
      • title = {Learning invariant Riemannian geometric representations using deep nets},
      • booktitle = {Proceedings of the IEEE International Conference on Computer Vision Workshops},
      • year = 2017,
      • pages = {1329--1338}
      • }
    •  Kuldeep Kulkarni, Suhas Lohit, Pavan Turaga, Ronan Kerviche and Amit Ashok, "Reconnet: Non-iterative reconstruction of images from compressively sensed measurements", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 449-458.
      BibTeX
      • @Inproceedings{kulkarni2016reconnet,
      • author = {Kulkarni, Kuldeep and Lohit, Suhas and Turaga, Pavan and Kerviche, Ronan and Ashok, Amit},
      • title = {Reconnet: Non-iterative reconstruction of images from compressively sensed measurements},
      • booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
      • year = 2016,
      • pages = {449--458}
      • }
    •  Suhas Lohit, Kuldeep Kulkarni and Pavan Turaga, "Direct inference on compressive measurements using convolutional neural networks", 2016 IEEE International Conference on Image Processing (ICIP), 2016, pp. 1913-1917.
      BibTeX
      • @Inproceedings{lohit2016direct,
      • author = {Lohit, Suhas and Kulkarni, Kuldeep and Turaga, Pavan},
      • title = {Direct inference on compressive measurements using convolutional neural networks},
      • booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
      • year = 2016,
      • pages = {1913--1917},
      • organization = {IEEE}
      • }
    •  Qiao Wang, Suhas Lohit, Meynard John Toledo, Matthew P Buman and Pavan Turaga, "A statistical estimation framework for energy expenditure of physical activities from a wrist-worn accelerometer", 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2016, pp. 2631-2635.
      BibTeX
      • @Inproceedings{wang2016statistical,
      • author = {Wang, Qiao and Lohit, Suhas and Toledo, Meynard John and Buman, Matthew P and Turaga, Pavan},
      • title = {A statistical estimation framework for energy expenditure of physical activities from a wrist-worn accelerometer},
      • booktitle = {2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)},
      • year = 2016,
      • pages = {2631--2635},
      • organization = {IEEE}
      • }
    •  Suhas Lohit, Kuldeep Kulkarni, Pavan Turaga, Jian Wang and Aswin C Sankaranarayanan, "Reconstruction-free inference on compressive measurements", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2015, pp. 16-24.
      BibTeX
      • @Inproceedings{lohit2015reconstruction,
      • author = {Lohit, Suhas and Kulkarni, Kuldeep and Turaga, Pavan and Wang, Jian and Sankaranarayanan, Aswin C},
      • title = {Reconstruction-free inference on compressive measurements},
      • booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
      • year = 2015,
      • pages = {16--24}
      • }
  • Software Downloads

  • MERL Issued Patents

    • Title: "Systems and Methods for Multi-Spectral Image Fusion Using Unrolled Projected Gradient Descent and Convolutinoal Neural Network"
      Inventors: Liu, Dehong; Lohit, Suhas; Mansour, Hassan; Boufounos, Petros T.
      Patent No.: 10,891,527
      Issue Date: Jan 12, 2021
    See All Patents for MERL