Krzysztof Kusnierek
Head of Department/Head of Research
(+47) 920 12 953
krzysztof.kusnierek@nibio.no
Place
Apelsvoll
Visiting address
Nylinna 226, 2849 Kapp
Abstract
Neural Radiance Fields (NeRF) have been widely adopted for reconstructing high-quality 3D scenes from 2D RGB images. However, achieving accurate 3D object segmentation within these reconstructed scenes remains challenging. Existing NeRF-based segmentation methods either rely on post-processing (SA3D), which produces noisy point clouds due to the absence of density field optimization, or employ joint training with additional segmentation heads (FruitNeRF), which can lead to suboptimal performance due to conflicting learning objectives. In this work, we propose InvNeRF-Seg (Input-substitution NeRF for Segmentation), a two-stage fine-tuning strategy for 3D object segmentation that preserves the original NeRF architecture and loss function entirely. We first train a standard NeRF on RGB images and then fine-tune it using 2D segmentation masks formatted as RGB-like inputs, without introducing any architectural modifications or additional loss functions. This input-substitution approach reshapes the density field to align with object regions while suppressing background density. We validate InvNeRF-Seg through comprehensive ablation studies examining the roles of density and color MLPs, loss function choices, and training strategies. Field density analysis reveals consistent semantic refinement: densities of object regions increase while background densities are suppressed. Experiments on synthetic fruit datasets and real-world soybean imagery demonstrate that InvNeRF-Seg produces cleaner 3D segmented point clouds compared to both SA3D and FruitNeRF, enabling more accurate downstream object counting. The method is further validated on a self-collected soybean dataset to demonstrate its applicability in real-world agricultural scenarios.
Authors
Henk Maessen Siv Mari Aurdal Tomasz Leszek Woznicki Krzysztof Kusnierek Trond Haraldsen Anita SønstebyAbstract
Bruk av trefiber vil redusere miljøbelastningen ved å erstatte energikrevende steinullproduksjon med et fornybart materiale basert på rest-råstoff fra norsk skogindustri. Ved endt bruk kan trefibermaterialet komposteres, brukes som jordforbedringsmiddel eller brennes for energigjenvinning, i motsetning til steinull som ender på deponi. Etter lovende resultater med Fibergrow® trefibermatter i tomat i 2022 og 2023, gjennomførte vi nye forsøk i 2025 for å bekrefte disse funnene.
Authors
Linghan Huang Tingxuan Zhuang Meiqi Zhang Syed Tahir Ata-Ul-Karim Kang Yu Krzysztof Kusnierek Wei Li Xiaojun Liu Yongchao Tian Yan Zhu Weixing Cao Qiang CaoAbstract
Early-season prediction of winter wheat yield and grain protein content is essential for guiding fertilizer and irrigation decisions and reducing uncertainty in variable agroecosystems, as yield affects profitability and quality affects market value and nutrition. Although multi-source data and both single-task learning (STL) and multi-task learning (MTL) are widely used for predicting grain yield and quality, the conditions under which each approach performs best remain poorly understood, especially when data availability, noise, and measurement or computational constraints vary. To address this gap, we conducted a three-year field experiment in Henan Province, China, compiling environmental, agronomic, and proximal-sensing variables across five growth stages. Seven subsets were constructed, including environmental, agronomic, sensor, and combined subsets, and STL/MTL variants of Multilayer Perceptron (MLP), Transformer, and Random Forest (RF) were benchmarked. SHapley Additive exPlanations (SHAP) analysis quantified feature- and stage-level contributions and guided construction of compact Top-K subsets for accuracy–efficiency trade-offs. Multi-source fusion substantially improved accuracy over single-source inputs, with the combined agronomic-sensor subset providing the best performance (yield R2 = 0.823; GPC R2 = 0.743). Under the current stage-aggregated multi-source representation, MLPs outperformed Transformers and RFs across configurations, indicating that compact nonlinear models were better suited to the present medium-dimensional tabular setting. MTL provided the greatest benefit with sparse feature sets or imbalanced predictive difficulty, whereas STL performed better when information was abundant and signals were strong. SHAP analysis showed that agronomic and sensor features associated with biomass accumulation, nitrogen status, water availability, and canopy light interception were key drivers of model predictions, particularly during erecting and early grain filling. These findings further show that the value of STL versus MTL depends on data-source composition and information richness, and that SHAP can be used not only for interpretation but also for reduced-feature subset design. Within the present plot-scale setting, this study therefore provides a decision-oriented framework for identifying both accuracy-oriented and efficiency-oriented configurations, with efficiency referring to feature parsimony, reduced input and preprocessing burden, and computational time for winter wheat yield and GPC.
Division of Food Production and Society
Gene2Bread: Building knowledge and exploiting technology to achieve high wheat self-sufficiency in Norway
The project aims to increase Norway’s self-sufficiency in milling wheat by gaining knowledge to improve wheat quality and increase its utilisation. Adequate protein content and a stable quality between seasons and within a season are prerequisites to increasing the share of Norwegian wheat in the flour blend.
Division of Environment and Natural Resources
Sinograin III: Smart agricultural technology and waste-made biochar for food security, reduction of greenhouse gas (GHG) emission, and bio-and circular economy
The Sinograin III project’s overall objective is to contribute to the UN SDGs by widely implementing precision agriculture technologies and application of “waste-to-value” biochar products to achieve sustainable food production with minimized GHG emission, improve soil fertility and promote green growth/zero waste in modern agriculture in China.