Research Projects
My research spans computer vision, machine learning, and their applications to science and conservation. See also Google Scholar for a full list of publications.
Fine-grained Recognition and Texture Understanding
A long-standing theme in my work is understanding visual categories that require fine-grained discrimination — species, materials, textures, and scenes. Early work established bilinear CNN models and deep filter banks that capture second-order statistics for texture and fine-grained recognition. More recently we have studied how vision-language models (VLMs) can be adapted for zero-shot and few-shot recognition, and how to improve transductive inference for fine-grained tasks.
- Bilinear CNNs for Fine-grained Visual Recognition, Tsung-Yu Lin, Aruni RoyChowdhury, Subhransu Maji, IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2017 | webpage
- Describing Textures using Natural Language, Chenyun Wu, Mikayla Timm, Subhransu Maji, European Conference on Computer Vision (ECCV), 2020
- Improved Zero-Shot Classification by Adapting VLMs with Text Descriptions, Oindrila Saha, Grant Van Horn, Subhransu Maji, Computer Vision and Pattern Recognition (CVPR), 2024
- Generate, Transduct, Adapt: Iterative Transduction with VLMs, Oindrila Saha, Logan Lawrence, Grant Van Horn, Subhransu Maji, International Conference on Computer Vision (ICCV), 2025
3D Shape Understanding and Synthesis
We develop methods for representing, reconstructing, segmenting, and generating 3D shapes. Early work on multi-view CNNs and SPLATNet established strong baselines for 3D shape recognition. Later work explored learning structured shape representations (CSGNet, ParSeNet), unsupervised shape generation from 2D views, and differentiable reconstruction. More recently we have extended this to 3D-aware image editing and novel view synthesis using neural radiance fields.
- Multi-view Convolutional Neural Networks for 3D Shape Recognition, Hang Su, Subhransu Maji, Evangelos Kalogerakis, Erik Learned-Miller, International Conference on Computer Vision (ICCV), 2015 | webpage
- CSGNet: Neural Shape Parser for Constructive Solid Geometry, Gopal Sharma, Rishabh Goyal, Difan Liu, Evangelos Kalogerakis, Subhransu Maji, Computer Vision and Pattern Recognition (CVPR), 2018
- LU-NeRF: Scene and Pose Estimation by Synchronizing Local Unposed NeRFs, Zezhou Cheng, Carlos Esteves, Varun Jampani, Abhishek Kar, Subhransu Maji, Ameesh Makadia, International Conference on Computer Vision (ICCV), 2023
- 3D Space as a Scratchpad for Editable Text-to-Image Generation, Oindrila Saha, Vojtech Krs, Radomir Mech, Subhransu Maji, Matheus Gadelha, Kevin James Blackburn-Matzen, Computer Vision and Pattern Recognition (CVPR), 2026
Few-shot, Meta-learning, and Self-supervised Learning
We study how to learn visual representations and models that generalize from few examples. This includes meta-learning algorithms for few-shot classification, task embedding methods that characterize dataset complexity, self-supervised and semi-supervised learning for fine-grained categories, and analyzing when self-supervision helps few-shot learning.
- Task2Vec: Task Embedding for Meta-Learning, Alessandro Achille, Michael Lam, Rahul Tewari, Avinash Ravichandran, Subhransu Maji, Charless Fowlkes, Stefano Soatto, Pietro Perona, International Conference on Computer Vision (ICCV), 2019 | code
- Meta-Learning with Differentiable Convex Optimization, Kwonjoon Lee, Subhransu Maji, Avinash Ravichandran, Stefano Soatto, Computer Vision and Pattern Recognition (CVPR), 2019 | code
- When Does Self-Supervision Improve Few-Shot Learning?, Jong-Chyi Su, Subhransu Maji, Bharath Hariharan, European Conference on Computer Vision (ECCV), 2020
- Task2Box: Box Embeddings for Modeling Asymmetric Task Relationships, Rangel Daroya, Aaron Sun, Subhransu Maji, Computer Vision and Pattern Recognition (CVPR), 2024
Ecological Remote Sensing
We develop computer vision and machine learning methods for monitoring the natural environment at scale using weather radar and satellite imagery. In collaboration with ecologists and hydrologists, we have built systems for detecting bird migration patterns and communal roosts in Doppler radar, estimating historical migration phenology, and monitoring rivers from space using high-resolution satellite data.
- MistNet: Measuring Historical Bird Migration Using Archived Weather Radar and CNNs, Tsung-Yu Lin, Kevin Winner, Garrett Bernstein, Abhay Mittal, Adriaan M. Dokter, Kyle G. Horton, et al., Subhransu Maji, Daniel Sheldon, Methods in Ecology and Evolution, 2019
- Detecting and Tracking Communal Bird Roosts in Weather Radar Data, Zezhou Cheng, Saadia Gabriel, Pankaj Bhambhani, Daniel Sheldon, Subhransu Maji, Andrew Laughlin, David Winkler, Association for the Advancement of Artificial Intelligence (AAAI), 2020
- RiverScope: High-Resolution River Masking Dataset, Rangel Daroya, Taylor Rowley, Jonathan Flores, et al., Colin Gleason, Subhransu Maji, Association for the Advancement of Artificial Intelligence (AAAI) (oral), 2026
- WildSAT: Learning Satellite Image Representations from Wildlife Observations, Rangel Daroya, Elijah Cole, Oisin Mac Aodha, Grant Van Horn, Subhransu Maji, International Conference on Computer Vision (ICCV), 2025
Biodiversity and Species Understanding
We collaborate with biodiversity informatics researchers to develop machine learning methods for understanding species distributions and identification at scale. This includes learning species range maps from community science observations combined with language, building large-scale sound datasets for bioacoustics monitoring, and benchmarking multimodal models on fine-grained species identification tasks.
- Combining Observational Data and Language for Species Range Estimation, Max Hamilton, Christian Lange, Elijah Cole, Alexander Shepard, Samuel Heinrich, Oisin Mac Aodha, Grant Van Horn, Subhransu Maji, Neural Information Processing Systems (NeurIPS), 2024
- Feedforward Few-shot Species Range Estimation, Christian Lange, Max Hamilton, Elijah Cole, Alexander Shepard, Samuel Heinrich, Angela Zhu, Subhransu Maji, Grant Van Horn, Oisin Mac Aodha, International Conference for Machine Learning (ICML), 2025
- The iNaturalist Sound Dataset, Mustafa Chasmai, Alexander Shepard, Subhransu Maji, Grant Van Horn, Neural Information Processing Systems (NeurIPS), 2024
Active Learning and Scalable Estimation
We develop statistically efficient methods for estimation and counting in large image collections, motivated by ecological and scientific applications where ground truth labels are expensive. This includes importance sampling approaches for counting objects in aerial surveys, human-in-the-loop visual re-identification for population size estimation, and active measurement methods for estimating statistics of scientific images at scale.
- DISCount: Counting in Large Image Collections with Detector-based Importance Sampling, Gustavo Perez, Subhransu Maji, Daniel Sheldon, Association for the Advancement of Artificial Intelligence (AAAI), 2024 — Best Paper, AI for Social Impact Track
- Human-in-the-Loop Visual Re-ID for Population Size Estimation, Gustavo Perez, Daniel Sheldon, Grant Van Horn, Subhransu Maji, European Conference on Computer Vision (ECCV), 2024
- Active Measurement: Efficient Estimation at Scale, Max Hamilton, Jinlin Lai, Wenlong Zhao, Subhransu Maji, Daniel Sheldon, Neural Information Processing Systems (NeurIPS), 2025
AI for Physical Sciences
We apply computer vision to problems in astronomy and materials science. In astronomy, we have developed deep learning methods for classifying star clusters in Hubble Space Telescope images and built AI-assisted tools for cataloging galaxies at scale. In materials science, we have developed 3D CNNs for predicting adsorption properties of nanoporous zeolites from their molecular structure.
- StarcNet: Machine Learning for Star Cluster Classification, Gustavo Perez, Matteo Messa, Daniela Calzetti, Subhransu Maji, Dooseok Jung, Angela Adamo, Mattia Sirressi, The Astrophysical Journal, 2021 — CICS Outstanding Synthesis Award
- ZeoNet: 3D CNNs for Predicting Adsorption in Nanoporous Zeolites, Yachan Liu, Gustavo Perez, Zezhou Cheng, Aaron Sun, Samuel Hoover, Wei Fan, Subhransu Maji, Peng Bai, Journal of Materials Chemistry A, 2023