José Daniel Hernández Sosa

Jose Sosa, Mohamed Aloulou, Danila Rukhovich et al.

Self-supervised pre-training has proven highly effective for many computer vision tasks, particularly when labelled data are scarce. In the context of Earth Observation (EO), foundation models and various other Vision Transformer (ViT)-based approaches have been successfully applied for transfer learning to downstream tasks. However, it remains unclear under which conditions pre-trained models offer significant advantages over training from scratch. In this study, we investigate the effectiveness of pre-training ViT-based Masked Autoencoders (MAE) for downstream EO tasks, focusing on reconstruction, segmentation, and classification. We consider two large ViT-based MAE pre-trained models: a foundation model (Prithvi) and SatMAE. We evaluate Prithvi on reconstruction and segmentation-based downstream tasks, and for SatMAE we assess its performance on a classification downstream task. Our findings suggest that pre-training is particularly beneficial when the fine-tuning task closely resembles the pre-training task, e.g. reconstruction. In contrast, for tasks such as segmentation or classification, training from scratch with specific hyperparameter adjustments proved to be equally or more effective.

Jose Sosa, Danila Rukhovich, Anis Kacem et al.

Multi-modal data in Earth Observation (EO) presents a huge opportunity for improving transfer learning capabilities when pre-training deep learning models. Unlike prior work that often overlooks multi-modal EO data, recent methods have started to include it, resulting in more effective pre-training strategies. However, existing approaches commonly face challenges in effectively transferring learning to downstream tasks where the structure of available data differs from that used during pre-training. This paper addresses this limitation by exploring a more flexible multi-modal, multi-task pre-training strategy for EO data. Specifically, we adopt a Multi-modal Multi-task Masked Autoencoder (MultiMAE) that we pre-train by reconstructing diverse input modalities, including spectral, elevation, and segmentation data. The pre-trained model demonstrates robust transfer learning capabilities, outperforming state-of-the-art methods on various EO datasets for classification and segmentation tasks. Our approach exhibits significant flexibility, handling diverse input configurations without requiring modality-specific pre-trained models. Code will be available at: https://github.com/josesosajs/multimae-meets-eo.

Jose Sosa, Dan Pineau, Arunkumar Rathinam et al.

Monocular 6-DoF pose estimation plays an important role in multiple spacecraft missions. Most existing pose estimation approaches rely on single images with static keypoint localisation, failing to exploit valuable temporal information inherent to space operations. In this work, we adapt a deep learning framework from human pose estimation to the spacecraft pose estimation domain that integrates motion-aware heatmaps and optical flow to capture motion dynamics. Our approach combines image features from a Vision Transformer (ViT) encoder with motion cues from a pre-trained optical flow model to localise 2D keypoints. Using the estimates, a Perspective-n-Point (PnP) solver recovers 6-DoF poses from known 2D-3D correspondences. We train and evaluate our method on the SPADES-RGB dataset and further assess its generalisation on real and synthetic data from the SPARK-2024 dataset. Overall, our approach demonstrates improved performance over single-image baselines in both 2D keypoint localisation and 6-DoF pose estimation. Furthermore, it shows promising generalisation capabilities when testing on different data distributions.