Post by Zhengpeng Feng

🚀 Thrilled to announce that my very first PhD paper has been accepted at #CVPR2026! 🎉 As I settle into the second year of my PhD journey at Cambridge, I am proud to share this major milestone from my first year's work. We are introducing TESSERA (Temporal Embeddings of Surface Spectra for Earth Representation and Analysis) 🌍🛰️ Working with satellite Earth-observation (EO) time series is notoriously challenging—optical and microwave data are often irregular due to orbital patterns and cloud cover. Traditional compositing methods try to fix this but end up losing critical temporal information (like vegetation phenology). In our work, we take a different path. TESSERA is a pixel-wise foundation model for multi-modal (Sentinel-1/2) EO time series. Instead of filtering out imperfect images, TESSERA learns robust, label-efficient embeddings from them. Even better, we advance the "Embeddings-as-Data" paradigm, using Quantization-Aware Training to compress 10m-resolution embeddings to int8, reducing storage needs by ~4x with near-zero performance drop. This makes planetary-scale analysis accessible and computationally feasible! 🙌 A massive thank you to my incredible team! This was a true team effort. I am deeply grateful for the brilliant minds: Clement Atzberger, Sadiq Jaffer, Jovana Knezevic, Madeline Lisaius, Markus Immitzer, James Ball, David Coomes, Anil Madhavapeddy, Andrew Blake, Srinivasan Keshav We truly built this together! If you are interested in Earth Observation and Foundation Models, please check out our work: 🔗 Paper: https://lnkd.in/dhz2d2wi 🌍 Embedding Access: https://lnkd.in/ejraTSc9 💻 Code & Weights: https://lnkd.in/ejNevdVZ Looking forward to sharing more updates and seeing everyone at CVPR in Denver! #ComputerVision #RemoteSensing #EarthObservation #Cambridge #MachineLearning #DeepLearning #PhD #PhDLife #CVPR2026