Explore the vast range of titles available.

Smallholder farms dominate in many parts of the world, including Sub-Saharan Africa. These systems are characterized by small, heterogeneous, and often indistinct field patterns, requiring a specialized methodology to map agricultural landcover. In this thesis, we developed a benchmark labeled data set of high-resolution satellite imagery of agricultural fields in South Africa. We presented a new approach to mapping agricultural fields, based on efficient extraction of a vast set of simple, highly correlated, and interdependent features, followed by a random forest classifier. The algorithm achieved similar high performance across agricultural types, including spectrally indistinct smallholder fields, and demonstrated the ability to generalize across large geographic areas. In sensitivity analyses, we determined multi-temporal images provided greater performance gains than the addition of multi-spectral bands. We also demonstrated how active learning can be incorporated in the algorithm to create smaller, more efficient training data sets, which reduced computational resources, minimized the need for humans to hand-label data, and boosted performance. We designed a patch-based uncertainty metric to drive the active learning framework, based on the regular grid of a crowdsourcing platform, and demonstrated how subject matter experts can be replaced with fleets of crowdsourcing workers. Our active learning algorithm achieved similar performance as an algorithm trained with randomly selected data, but with 62% less data samples. This thesis furthers the goal of providing accurate agricultural landcover maps, at a scale that is relevant for the dominant smallholder class. Accurate maps are crucial for monitoring and promoting agricultural production. Furthermore, improved agricultural landcover maps will aid a host of other applications, including landcover change assessments, cadastral surveys to strengthen smallholder land rights, and constraints for crop modeling and famine prediction.

Sub-Saharan Africa and other developing regions of the world are dominated by smallholder farms, which are characterized by small, heterogeneous, and often indistinct field patterns. In previous work, we developed an algorithm for mapping both smallholder and commercial agricultural fields that includes efficient extraction of a vast set of simple, highly correlated, and interdependent features, followed by a random forest classifier. In this paper, we demonstrated how active learning can be incorporated in the algorithm to create smaller, more efficient training data sets, which reduced computational resources, minimized the need for humans to hand-label data, and boosted performance. We designed a patch-based uncertainty metric to drive the active learning framework, based on the regular grid of a crowdsourcing platform, and demonstrated how subject matter experts can be replaced with fleets of crowdsourcing workers. Our active learning algorithm achieved similar performance as an algorithm trained with randomly selected data, but with 62% less data samples.

Smallholder farms dominate in many parts of the world, particularly Sub-Saharan Africa. These systems are characterized by small, heterogeneous, and often indistinct field patterns, requiring a specialized methodology to map agricultural land cover. Using a variety of sites in South Africa, we present a new approach to mapping agricultural fields, based on efficient extraction of a vast set of simple, highly correlated, and interdependent features, followed by a random forest classifier. We achieved similar high performance across agricultural types, including the spectrally indistinct smallholder fields as well as the more easily distinguishable commercial fields, and demonstrated the ability to generalize performance across large geographic areas. In sensitivity analyses, we determined multi-temporal information provided greater gains in accuracy than multi-spectral information.

Accurate landcover maps are fundamental to understanding socio-economic and environmental patterns and processes, but existing datasets contain substantial errors. Crowdsourcing map creation may substantially improve accuracy, particularly for discrete cover types, but the quality and representativeness of crowdsourced data is hard to verify. We present an open-sourced platform, DIYlandcover, that serves representative samples of high resolution imagery to an online job market, where workers delineate individual landcover features of interest. Worker mapping skill is frequently assessed, providing estimates of overall map accuracy and a basis for performance-based payments. A trial of DIYlandcover showed that novice workers delineated South African cropland with 91% accuracy, exceeding the accuracy of current generation global landcover products, while capturing important geometric data. A scaling-up assessment suggests the possibility of developing an Africa-wide vector-based dataset of croplands for $2-3 million within 1.2-3.8 years. DIYlandcover can be readily adapted to map other discrete cover types.

Generating realistic 3D scenes is an area of growing interest in computer vision and robotics. However, creating high-quality, diverse synthetic 3D content often requires expert intervention, making it costly and complex. Recently, efforts to automate this process with learning techniques, particularly diffusion models, have shown significant improvements in tasks like furniture rearrangement. However, applying diffusion models to floor-conditioned indoor scene synthesis remains under-explored. This task is especially challenging as it requires arranging objects in continuous space while selecting from discrete object categories, posing unique difficulties for conventional diffusion methods. To bridge this gap, we present MiDiffusion, a novel mixed discrete-continuous diffusion model designed to synthesize plausible 3D indoor scenes given a floor plan and pre-arranged objects. We represent a scene layout by a 2D floor plan and a set of objects, each defined by category, location, size, and orientation. Our approach uniquely applies structured corruption across mixed discrete semantic and continuous geometric domains, resulting in a better-conditioned problem for denoising. Evaluated on the 3D-FRONT dataset, MiDiffusion outperforms state-of-the-art autoregressive and diffusion models in floor-conditioned 3D scene synthesis. Additionally, it effectively handles partial object constraints via a corruption-and-masking strategy without task-specific training, demonstrating advantages in scene completion and furniture arrangement tasks.

Vision is essential for human navigation. The World Health Organization (WHO) estimates that 43.3 million people were blind in 2020, and this number is projected to reach 61 million by 2050. Modern scene understanding models could empower these people by assisting them with navigation, obstacle avoidance and visual recognition capabilities. The research community needs high quality datasets for both training and evaluation to build these systems. While datasets for autonomous vehicles are abundant, there is a critical gap in datasets tailored for outdoor human navigation. This gap poses a major obstacle to the development of computer vision based Assistive Technologies. To overcome this obstacle, we present SANPO, a large-scale egocentric video dataset designed for dense prediction in outdoor human navigation environments. SANPO contains 701 stereo videos of 30+ seconds captured in diverse real-world outdoor environments across four geographic locations in the USA. Every frame has a high resolution depth map and 112K frames were annotated with temporally consistent dense video panoptic segmentation labels. The dataset also includes 1961 high-quality synthetic videos with pixel accurate depth and panoptic segmentation annotations to balance the noisy real world annotations with the high precision synthetic annotations. SANPO is already publicly available and is being used by mobile applications like Project Guideline to train mobile models that help low-vision users go running outdoors independently. To preserve anonymization during peer review, we will provide a link to our dataset upon acceptance. SANPO is available here: https://google-research-datasets.github.io/sanpo_dataset/

Dense prediction tasks, such as semantic segmentation, depth estimation, and surface normal prediction, can be easily formulated as per-pixel classification (discrete outputs) or regression (continuous outputs). This per-pixel prediction paradigm has remained popular due to the prevalence of fully convolutional networks. However, on the recent frontier of segmentation task, the community has been witnessing a shift of paradigm from per-pixel prediction to cluster-prediction with the emergence of transformer architectures, particularly the mask transformers, which directly predicts a label for a mask instead of a pixel. Despite this shift, methods based on the per-pixel prediction paradigm still dominate the benchmarks on the other dense prediction tasks that require continuous outputs, such as depth estimation and surface normal prediction. Motivated by the success of DORN and AdaBins in depth estimation, achieved by discretizing the continuous output space, we propose to generalize the cluster-prediction based method to general dense prediction tasks. This allows us to unify dense prediction tasks with the mask transformer framework. Remarkably, the resulting model PolyMaX demonstrates state-of-the-art performance on three benchmarks of NYUD-v2 dataset. We hope our simple yet effective design can inspire more research on exploiting mask transformers for more dense prediction tasks. Code and model will be made available.

Crop model-specific biases are a key uncertainty affecting our understanding of climate change impacts to agriculture. There is increasing research focus on intermodel variation, but comparisons between mechanistic (MMs) and empirical models (EMs) are rare despite both being used widely in this field. We combined MMs and EMs to project future (2055) changes in the potential distribution (suitability) and productivity of maize and spring wheat in South Africa under 18 downscaled climate scenarios (9 models run under 2 emissions scenarios). EMs projected larger yield losses or smaller gains than MMs. The EMs' median-projected maize and wheat yield changes were 3.6% and 6.2%, respectively, compared to 6.5% and 15.2% for the MM. The EM projected a 10% reduction in the potential maize growing area, where the MM projected a 9% gain. Both models showed increases in the potential spring wheat production region (EM = 48%, MM = 20%), but these results were more equivocal because both models (particularly the EM) substantially overestimated the extent of current suitability. The substantial water-use efficiency gains simulated by the MMs under elevated CO2 accounted for much of the EMMM difference, but EMs may have more accurately represented crop temperature sensitivities. Our results align with earlier studies showing that EMs may show larger climate change losses than MMs. Crop forecasting efforts should expand to include EMMM comparisons to provide a fuller picture of crop-climate response uncertainties.