Search for: All records

Amodal segmentation is an image-based algorithm that aims to predict masks for both visible and occluded parts of objects. Existing methods typically rely on supervised learning with annotated amodal masks or synthetic data. The effectiveness of these methods relies heavily on the quality of the datasets. This dependence can unintentionally restrict their generalization capabilities due to insufficient diversity and size. Although existing zero-shot methods perform well on their reported datasets, their performance does not necessarily transfer to other datasets. We propose a tuning-free approach that re-purposes diffusion-based inpainting foundation models for amodal segmentation. Our approach is motivated by the “occlusion-free bias” of inpainting models, i.e., the inpainted objects tend to be complete and without occlusions. We reconstruct the occluded regions of an object via inpainting and then apply segmentation, all without additional training or fine-tuning. Experiments on five datasets, three previously unreported, demonstrate the generalizability of our approach. On average, our approach achieves 5.3% more accurate masks in mIoU compared to the publicly available state-of-the-art, pix2gestalt. 

Finding where the trees are in a city and monitoring any changes are essential for sustainable urban management. Historically, urban forests are mainly inventoried via manual processes often limited to public lands. Leveraging advances in computing, we present a novel generative artificial intelligence (AI) method along with a first-ever national-scale dataset, to automatically localize trees in cities across the nation using satellite imagery. Our monitoring approach is fully automated and can be completed for 330 U.S. cities within less than a day of computing, enabling actionable knowledge of changes in urban trees and supporting sustainable development decisions. We successfully localized and counted over 278 million trees, achieving an average tree count accuracy of 92.5% and spatial accuracy of 1.5m for 2024–2025. Our computational approach allows for novel nationwide analysis to be performed. For example, we can localize approximately 117 million trees on private lands and 161 million on public lands. Further, we show and quantify that urban tree distribution exhibits strong spatial disparity, with low-income communities having substantially fewer trees and less canopy cover than others. In addition, we compare tree count and layouts before and after multiple major events (e.g., major fires and destructive weather phenomena). Overall, our approach enhances computational urban planning, including weather and extreme event forecasting, for the development of sustainable cities. 

Model immunization is an emerging direction that aims to mitigate the potential risk of misuse associated with open-sourced models and advancing adaptation methods. The idea is to make the released models' weights difficult to fine-tune on certain harmful applications, hence the name immunized. Recent work on model immunization focuses on the single-concept setting. However, in real-world situations, models need to be immunized against multiple concepts. To address this gap, we propose an immunization algorithm that, simultaneously, learns a single difficult initialization for adaptation methods over a set of concepts. We achieve this by incorporating a differentiable merging layer that combines a set of model weights adapted over multiple concepts.In our experiments, we demonstrate the effectiveness of multi-concept immunization by generalizing prior work's experiment setup of re-learning and personalization adaptation to multiple concepts. 

Abstract The task of crafting procedural programs capable of generating structurally valid 3D shapes easily and intuitively remains an elusive goal in computer vision and graphics. Within the graphics community, generating procedural 3D models has shifted to using node graph systems. They allow the artist to create complex shapes and animations through visual programming. Being a high‐level design tool, they made procedural 3D modelling more accessible. However, crafting those node graphs demands expertise and training. We present GeoCode, a novel framework designed to extend an existing node graph system and significantly lower the bar for the creation of new procedural 3D shape programs. Our approach meticulously balances expressiveness and generalization for part‐based shapes. We propose a curated set of new geometric building blocks that are expressive and reusable across domains. We showcase three innovative and expressive programs developed through our technique and geometric building blocks. Our programs enforce intricate rules, empowering users to execute intuitive high‐level parameter edits that seamlessly propagate throughout the entire shape at a lower level while maintaining its validity. To evaluate the user‐friendliness of our geometric building blocks among non‐experts, we conduct a user study that demonstrates their ease of use and highlights their applicability across diverse domains. Empirical evidence shows the superior accuracy of GeoCode in inferring and recovering 3D shapes compared to an existing competitor. Furthermore, our method demonstrates superior expressiveness compared to alternatives that utilize coarse primitives. Notably, we illustrate the ability to execute controllable local and global shape manipulations. Our code, programs, datasets and Blender add‐on are available athttps://github.com/threedle/GeoCode. 

We present a novel approach to perform instance segmentation and counting for densely packed self-similar trees using a top-view RGB image sequence. We propose a solution that leverages pixel content, shape, and self-occlusion. First, we perform an initial over-segmentation of the image sequence and aggregate structural characteristics into a contour graph with temporal information incorporated. Second, using a graph convolutional network and its inherent local messaging passing abilities, we merge adjacent tree crown patches into a final set of tree crowns. Per various studies and comparisons, our method is superior to all prior methods and results in high-accuracy instance segmentation and counting despite the trees being tightly packed. Finally, we provide various forest image sequence datasets suitable for subsequent benchmarking and evaluation captured at different altitudes and leaf conditions.