Max Ehrlich

I am a Ph.D. student in Computer Science at the University of Maryland and a research assistant in the Institute for Advanced Computer Studies. My adviser is Dr. Larry Davis and I am co-advised by Dr. Abhinav Shrivastava. Before starting my Ph.D. I was a scientist at SRI International's Center for Vision Technologies. I received a B.S. in Computer Science from Rutgers University and an M.S. in Computer Science from Stevens Institute of Technology. My current research area is compressed methods for deep learning. This includes how to accelerate learning using compressed image representations, how to better use compressed data in the machine learning setting and control its accuracy penalty, and how to compress images/videos better using deep learning.

I am a student member of the Association for the Advancement of Artificial Intelligence (AAAI).


Contact me by email at

maxehr {at} umiacs {dot} umd {dot} edu
or drop by my office at IRB 3116.

Send regular mail to

        Max Ehrlich
        c/o Department of Computer Science
        5109 Brendan Iribe Center for Computer Science and Engineering
        8125 Paint Branch Drive
        College Park, MD 20742


  • CRAB: Compression Robustness Analysis Benchmark Paper (arXiv) Cite It! - The most comprehensive study of the effect of JPEG compression to date! The CRAB system allows for fast, easy, and consistent benchmarking of deep learning methods when their inputs are JPEG compressed, as well as how they behave under various JPEG mitigation techniques including a new one we developed that is entirely self-supervised. We used CRAB to benchmark 20 commonly used models across three tasks: classification, detection, and segmentation (instance and semantic). Stay tuned for the CRAB code release, which will allow researchers to benchmark their own models and submit the results for inclusion into the study as well as the study website detailing our findings. In the meantime, check out our preprint on arXiv ( which contains details of the study as well as the complete results.
  • Quantization Guided JPEG Artifact Correction Project Page Code Talk Slides Talk Video Paper (arXiv) Cite It! - We develop a novel method for JPEG artifact correction that solves three major problems left open in prior works:
    1. Prior works train an ensemble of models, one for each JPEG quality. We use a single network parameterized by the JPEG quantization matrix.
    2. Prior works deal with grayscale images only, with the assumption that their models can be applied channel-wise. We show that single-channel networks have trouble generalizing and design a network for color correction.
    3. Prior works focus on CNN regression which causes blurry and textureless results. We introduce a novel GAN loss that includes an explicit texture restoring term, this yields a more realistic result.
    Our method achieves state-of-the-art results on color artifact correction. The paper was published in the proceedings of the European Conference on Computer Vision. I strongly recommend reading the arXiv version which includes the appendices.
  • JPEG Domain Residual Networks Notes Colab Code Poster Paper Cite It! - In this work we develop the popular Residual Network architecture in the JPEG domain. Our goal is to produce a formulation which gives a result that is as possible to the spatial domain network, but which can operate on compressed JPEG images. Our formulation is generic and has applicability outside of classification objective that we show as an example. We show a notable performance increase by processing in the JPEG domain. This work was funded by Facebook Inc. and published in the proceedings of the International Conference on Computer Vision 2019.
  • CORE3D IARPA - The CORE3D program aims to develop technology that generates, in an automated way, accurate 3D object models with real physical properties, from multiple data sources including commercial satellite panchromatic and multi-spectral imagery for global coverage, and airborne imagery and Geographic Information System (GIS) vector data, where available, for improved resolution and fidelity.
  • Squad-X DARPA - DARPA's Squad X Core Technologies (SXCT) program aims to develop novel technologies that could be integrated into user-friendly systems that would extend squad awareness and engagement capabilities without imposing physical and cognitive burdens. The goal is to speed the development of new, lightweight, integrated systems that provide infantry squads unprecedented awareness, adaptability and flexibility in complex environments, and enable dismounted Soldiers and Marines to more intuitively understand and control their complex mission environments.

Papers and Patents

Google Scholar dblp Semantic Scholar

Total Publications: Total Citations: h-index:

Loading citation data from Semantic Scholar
  • Max Ehrlich, Larry Davis, Ser-Nam Lim, and Abhinav Shrivastava. "Analysing and Mitigating Compression Defects in Deep Learning" arXiv preprint arXiv:2011.08932, 2020. arXiv Cite It!
  • Max Ehrlich, Larry Davis, Ser-Nam Lim, and Abhinav Shrivastava. "Quantization Guided JPEG Artifact Correction." In Proceedings of the European Conference on Computer Vision, 2020. arXiv ECVA Cite It!
  • Arthita Ghosh, Max Ehrlich, Larry Davis, and Rama Chellappa. "Unsupervised Super-Resolution of Satellite Imagery for High Fidelity Material Label Transfer." In IEEE International Geoscience and Remote Sensing Symposium, pp. 5144-5147. IEEE, 2019. IEEE Cite It!
  • Max Ehrlich, and Larry S. Davis. "Deep Residual Learning in the JPEG Transform Domain." In Proceedings of the IEEE International Conference on Computer Vision, pp. 3484-3493. 2019. arXiv CVF Cite It!
  • Mohamed R. Amer, Timothy J. Shields, Amir Tamrakar, Max Ehrlich, and Timur Almaev. "Deep Multi-Task Representation Learning." U.S. Patent Application 16/085,859, filed January 31, 2019. Google Cite It!
  • Arthita Ghosh, Max Ehrlich, Sohil Shah, Larry Davis, and Rama Chellappa. "Stacked U-Nets for Ground Material Segmentation in Remote Sensing Imagery." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 257-261. 2018. CVF Cite It!
  • Timothy J. Shields, Mohamed R. Amer, Max Ehrlich, Amir Tamrakar. "Action-Affect-Gender Classification using Multi-Task Representation Learning." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 2249-2258. 2017. CVF Cite It!
  • Max Ehrlich, and Philippos Mordohai. "Discriminative Hand Localization in Depth Images." In Proceedings of the IEEE Symposium on 3D User Interfaces, pp. 239-240. 2016. IEEE Direct Cite It!
  • Max Ehrlich, Timothy J. Shields, Timur Almaev, and Mohamed R. Amer. "Facial Attributes Classification using Multi-Task Representation Learning." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 47-55. 2016. CVF Cite It!
  • Max Ehrlich. "Discriminative Hand Tracking from Depth Images." Master's Thesis, Stevens Institute of Technology. 2015 Direct

Open Source

Github Gitlab

In general I am more active on gitlab than on github. All of my personal projects go on gitlab, as it is a truly open source solution unlike github which is closed source. I also prefer the tooling and UI on gitlab. My contributions to other projects are usually done on github since the majority of projects are currently using it.

Unless there is a company which owns the code I write, all my projects are open sourced under the MIT license. In general this means you can use my code for whatever purpose you want including commercial, I only ask that you credit me as the author. I usually make paper code public upon acceptance of the publication to a peer-reviewed venue. Here are a few highlights:

  • Linux ASCO Patch I currently maintain a patch and automatic .deb build system for the Linux kernel with ACS Override capability. If you need to bypass IOMMU groups for whatever reason and you don't have native ACS (and you use Ubuntu Linux), then you might be interested in my packages.
  • torchjpeg The goal of this library is to lower the barrier-to-entry for JPEG related deep learning tasks. The library includes functions to extract and write coefficients to files as well as manipulations on those coefficients (like quantization, resampling, and pixel transformations). It is the product of two years of research code with multiple published papers. It is available on PyPI. Also see the repository and documentation.