I graduated from the University of Electronic Science and Technology of China (UESTC) with a Bachelor of Engineering in Spatial Informatics and Digitalized Technology. After that, I joined the Energy and Environment Group (EEG) in the Department of Computer Science and Technology, University of Cambridge, as an Undergraduate Research Assistant under the supervision of my advisor, Srinivasan Keshav. Now, I am a PhD student under his supervision.
My research interest lies at the intersection of AI and Earth Science. I am particularly interested in developing Self-Supervised Learning (SSL) algorithms for remote sensing imagery.
I am also the co-founder and president of the China Artificial Intelligence Association (Cambridge). We are dedicated to creating a platform for AI communication that integrates academia and industry, connecting universities in China and the UK to build a cross-national AI talent ecosystem.
My main work at present is developing a remote sensing foundation model called TESSERA. We have recently released TESSERA v2, together with its code and model weights. We have found some exciting results.
We would like to express our gratitude to Isambard-AI, the UK’s most powerful AI supercomputer, and DAWN, the fastest artificial intelligence supercomputer at Cambridge, for their generous support in this project. We also acknowledge the support from NVIDIA, AMD, Vultr, Microsoft AI For Good Lab, dClimate, and Amazon Web Services (AWS). This work would not have been possible without their computational resources and technical assistance.
🔥 News
- 2026.07: 🎉🎉 I gave an oral presentation on TESSERA at the AI for Good Global Summit in Geneva, Switzerland.
- 2026.07: 🎉🎉 We released TESSERA v2! The preprint, the code, and the distilled model weights are all publicly available. We ran 395 pretraining runs on 1024 Nvidia H100 GPUs to work out how pixel-wise Earth foundation models scale.
- 2026.07: 🎉🎉 Our paper “Geospatial foundation models enable data-efficient tree species mapping in temperate mountain forests” was published in Science of Remote Sensing.
- 2026.06: 🎉🎉 TESSERA was featured by the European Space Agency (ESA): “Tessera AI model offers accessible way to view Earth”.
- 2026.06: 🎉🎉 TESSERA made its CVPR 2026 debut in Denver.
- 2026.06: 🎉🎉 Fresh blood for TESSERA! Two postdoctoral researchers and an assistant professor joined the team — welcome Silja Sormunen, Jingtao Li, and Kyle Gao.
- 2026.02: 🎉🎉 Our TESSERA paper was accepted by CVPR 2026.
- 2026.01: 🎉🎉 I was honored to be invited to the AI for Good Global Summit, held in Geneva, Switzerland, in July 2026, to present TESSERA-related works.
- 2025.12: 🎉🎉 I was honored to be invited by IEEE GRSS to give a talk about TESSERA. Information and recording can be found here.
- 2025.11: 🎉🎉 I was invited as a keynote speaker to the Open-Earth-Monitor Global Workshop 2026, which will be held in Barcelona, Spain, from 7–9 October 2026. The event is co-hosted by CREAF and the OpenGeoHub Foundation under the Open-Earth-Monitor-Cyberinfrastructure (OEMC) project.
- 2025.09: 🎉🎉 I was honored to participate in the 13th Heidelberg Laureate Forum as a young researcher and presented the TESSERA work (Heidelberg Laureate Forum).
- 2025.07: 🎉🎉 We have open-sourced the inference code for TESSERA and the code for using embeddings in downstream tasks (GeoTessera). We will be rolling out a global 10m resolution annual embedding product in the coming months.
Show older newsHide older news
- 2025.06: 🎉🎉 We have released the preprint of our latest work, TESSERA. We trained a 14B model to extract time-series spectral features from satellite images.
- 2025.03: 🎉🎉 Our app, GreenLens, won the Better Future Award from the Cambridge Ring!
- 2025.02: 🎉🎉 We published “SPREAD: A large-scale, high-fidelity synthetic dataset for multiple forest vision tasks” in Ecological Informatics. We used Unreal Engine 5 to create the most realistic synthetic data for 2D and 3D vision tasks in forests.
- 2024.10: 🎉🎉 I am thrilled to be fully funded to join the EEG, Department of Computer Science and Technology, University of Cambridge, as a PhD student.
- 2024.09: 🎉🎉 “An app for tree trunk diameter estimation from coarse optical depth maps,” was accepted by Ecological Informatics.
📝 Publications
TESSERA v2: Scaling Pixel-wise Earth Foundation Models
[Tessera]
Zhengpeng Feng, Sadiq Jaffer, Ira Shokar, Jovana Knezevic, Mark Elvers, Clement Atzberger, Robin Young, Aneesh Naik, Niall Robinson, Andrew Blake, David Coomes, Anil Madhavapeddy, Srinivasan Keshav
- The largest controlled scaling study for Earth observation to date: 395 pretraining runs on 1024 Nvidia H100 GPUs.
- Pretraining loss barely predicts downstream performance, so selecting models by loss wastes compute.
- Our distilled 21M-parameter TESSERA v2-1B-M outperforms all open and proprietary models tested.
- Matryoshka embeddings: a 16-dimensional prefix keeps 92% of the full performance at 1/8 of the storage.
TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis
[Tessera]
Zhengpeng Feng, Clement Atzberger, Sadiq Jaffer, Jovana Knezevic, Silja Sormunen, Robin Young, Madeline C. Lisaius, Markus Immitzer, Toby Jackson, James Ball, David A. Coomes, Anil Madhavapeddy, Andrew Blake, Srinivasan Keshav
- We trained a 1.4B model to extract time-series spectral features from satellite images.
- We are releasing 2.5PB of global 10m-resolution earth embeddings from 2017-2025.
- Researchers from over 200 groups worldwide are using TESSERA.
- Accepted by CVPR 2026.
Applications of the TESSERA Geospatial Foundation Model to Diverse Environmental Mapping Tasks
Zhengpeng Feng, Clement Atzberger, Sadiq Jaffer, Jovana Knezevic, Silja Sormunen, Robin Young, Madeline C. Lisaius, Markus Immitzer, Toby Jackson, James Ball, David A. Coomes, Anil Madhavapeddy, Andrew Blake, Srinivasan Keshav
- Under review at Remote Sensing of Environment (RSE).
- Extension of the CVPR 2026 TESSERA paper with additional downstream tasks and more detailed analysis.
James G.C. Ball, Jana Annika Wicklein, Zhengpeng Feng, Jovana Knezevic, Sadiq Jaffer, Anil Madhavapeddy, Clement Atzberger, Michele Dalponte, David A. Coomes
- We evaluate two geospatial foundation model embeddings, TESSERA and AlphaEarth, for mapping 18 tree species and groups in a demanding mountain landscape (Trentino, Italy).
- The embeddings consistently beat conventional Sentinel-1+2 composites (weighted F1 = 0.83 vs. 0.80), approaching saturation with only 5% of the training parcels and organising species into ecologically meaningful groupings.
- Realising this advantage requires a nonlinear classifier, and training with parcel-level species proportions as soft labels improves minority-species discrimination.
- Temporal transfer across years remains the bottleneck, shifting the challenge from feature engineering toward the availability, quality, and temporal alignment of reference data.
SPREAD: A large-scale, high-fidelity synthetic dataset for multiple forest vision tasks
Zhengpeng Feng, Yihang She, Srinivasan Keshav
- We used Unreal Engine 5 to create the super realistic synthetic data for 2D and 3D vision tasks in forests.
An app for tree trunk diameter estimation from coarse optical depth maps
Zhengpeng Feng, Mingyue Xie, Amelia Holcomb, Srinivasan Keshav
- An app for fast in-situ tree trunk diameter estimation.
- The app won the Cambridge Ring Better Future Award!
- More about the app
🎖 Honors and Awards
- 2025.09 Heidelberg Laureate Forum Young Researcher
- 2025.03 Cambridge Ring Better Future Award
- 2024.06 Robert Sansom Studentship
- 2022.12 The Most Outstanding Students Award of UESTC (top 10)
- 2022.11 Gratitude Scholarship for Chinese Modern Scientists
- 2022.10 National Scholarship
📖 Educations
- 2024.10 - Present, PhD in Computer Science, Department of Computer Science and Technology, University of Cambridge.
- 2023.07 - 2024.07, Research Assistant, Department of Computer Science and Technology, University of Cambridge.
- 2019.09 - 2023.06, B.Eng. in Spatial Informatics and Digitalized Technology, University of Electronic Science and Technology of China (UESTC).
💬 Invited Talks
-
2026.07, TESSERA, AI for Good Global Summit, Geneva, Switzerland (oral presentation). [link] - 2026.07, TESSERA v2, Photogrammetry and Remote Sensing (PRS) Group, ETH Zurich.
-
2025.12, TESSERA: Precomputed Fair Global Pixel Embeddings for Earth Representation and Analysis, IEEE GRSS. [link] -
2025.12, TESSERA, RISE. [video] -
2025.10, TESSERA, GEDI Group, University of Maryland. [link] -
2024.06, An App for Tree Trunk Diameter Estimation from Coarse Optical Depth Maps. [video]
💻 Supervisions
- 2025, 2026, Part IB CST Artificial Intelligence, University of Cambridge.