Engineering topics
Publications and source records attributed to Potnis, Abhishek.
The Oak Ridge National Laboratory (ORNL) team has successfully innovated geospatial data feature mapping and postprocessing workflows to support the USA Structures project supported by FEMA. This technical report highlights significant updates, improvements, and innovations for the USA Structures dataset, focusing on nine priority states.
While the pretraining of Foundation Models (FMs) for remote sensing (RS) imagery is on the rise, models remain restricted to a few hundred million parameters. Scaling models to billions of parameters has been shown to yield unprecedented benefits including emergent abilities, but requires data scaling and computing resources typically not available outside industry R&D labs. In this work, we pair high-performance computing resources including Frontier supercomputer, America's first exascale system, and high-resolution optical RS data to pretrain billion-scale FMs. Our study assesses performance of different pretrained variants of vision Transformers across image classification, semantic segmentation and object detection benchmarks, which highlight the importance of data scaling for effective model scaling. Moreover, we discuss construction of a novel TIU pretraining dataset, model initialization, with data and pretrained models intended for public release. By discussing technical challenges and details often lacking in the related literature, this work is intended to offer best practices to the geospatial community toward efficient training and benchmarking of larger FMs.
Computer vision algorithms are increasingly leveraged to accelerate geospatial analysis for disaster response and recovery. As the diversity of remote sensing imagery grows with optical, SAR, and other modalities, a perquisite for analytics is cross-modal image registration. There is a high potential to harness computer vision for this pre-processing requirement toward enabling downstream analytics such as heterogeneous change detection, automated feature extraction, and data fusion. Advancement in these areas has the potential to simplify data wrangling tasks and further accelerate disaster response timelines. The SpaceNet 9 challenge (launching in mid-2024) focuses on addressing the cross-modal image registration problem and demonstrating the utility of such modules on earthquake impacted scenarios. This paper describes the motivation for the SpaceNet 9 and provides a first overview of the dataset, the baseline algorithm, and implications for seeking cross-modal image registration in Earth observation. Code is available at https://github.com/SpaceNetChallenge/SpaceNet9.
The design of a pretraining dataset is emerging as a critical component for the generality of foundation models. In the remote sensing realm, large volumes of imagery and benchmark datasets exist that can be leveraged to pretrain foundation models, however using this imagery in absence of a well-crafted sampling strategy is inefficient and has the potential to create biased and less generalizable models. Here, we provide a discussion and vision for the curation and assessment of pretraining datasets for remote sensing geospatial foundation models. We highlight the importance of geographic, temporal, and image acquisition diversity and review possible strategies to enable such diversity at global scale. In addition to these characteristics, support for various spatial-temporal pretext tasks within the dataset is also critical. Ultimately, our primary objective is to place emphasis on and draw attention to the data curation stage of the foundation model development pipeline. By doing so, we think it is possible to reduce biases of geospatial foundation models, as well as enable broader generalization to downstream remote sensing tasks and applications.
As AI workloads increase in scope, generalization capability becomes challenging for small task-specific models and their demand for large amounts of labeled training samples increases. On the contrary, Foundation Models (FMs) are trained with internet-scale unlabeled data via self-supervised learning and have been shown to adapt to various tasks with minimal fine-tuning. Although large FMs have demonstrated significant impact in natural language processing and computer vision, efforts toward FMs for geospatial applications have been restricted to smaller size models, as pretraining larger models requires very large computing resources equipped with state-of-the-art hardware accelerators. Current satellite constellations collect 100+TBs of data a day, resulting in images that are billions of pixels and multimodal in nature. Such geospatial data poses unique challenges opening up new opportunities to develop FMs. We investigate billion scale FMs and HPC training profiles for geospatial applications by pretraining on publicly available data. We studied from end-to-end the performance and impact in the solution by scaling the model size. Our larger 3B parameter size model achieves up to 30% improvement in top1 scene classification accuracy when comparing a 100M parameter model. Moreover, we detail performance experiments on the Frontier supercomputer, America's first exascale system, where we study different model and data parallel approaches using PyTorch's Fully Sharded Data Parallel library. Specifically, we study variants of the Vision Transformer architecture (ViT), conducting performance analysis for ViT models with size up to 15B parameters. By discussing throughput and performance bottlenecks under different parallelism configurations, we offer insights on how to leverage such leadership-class HPC resources when developing large models for geospatial imagery applications.