You can think of 3DTWINZ TRAINBOW™ as a gym for machine learning systems. Each of our synthetic dataset pipeline datasets and modules is comparable to an exercise machine for a specific body part. Where your ML system is weak in - let's say accurately classifying one specific tree species with remote sensing (on aerial or satellite images or first person photos) or we can improve classification rates by adding high quality synthetic training data created with the 3DTWINZ TRAINBOW™ multi spectral synthetic dataset pipeline.
Klick here to learn more about 3DTWINZ™ in general.
The 3DTWINZ TRAINBOW™ Pipeline Framework allows you to improve your ML model results by on demand training it with 3d multispectral synthetic datasets tailored to address the weaknesses of your specific models. While Currently 3DTWINZ TRAINBOW™ is specialized on biotopes/habitats especially featuring forests, it can be adapted to any real world environment, whether natural/biological or artificial like buildings, machinery etc. The question we have to figure out for your project is: can 3DTWINZ TRAINBOW™ be used in an economically viable way for your special requirements?
Nov-2011 Machine learning experiments with distributed ML architectures like the Encog 3D Neural Pilot
Oct-2020 Deep dive into the status quo of current 3d synthetic datasets
Nov-2022 until Mar-2023 BIO.KI.S.S. Proof of Concept (PoC) (*BIOtoptypenanalyse-KI-Satellitenbild-Sampler) for DB Netz AG, Project management Joerg Osarek, with Fraunhofer IIS and Live-EO as AI service providers: technical proof of identifying biotopes/habitats on aerial and satellite images using RGB and NIR (Near Infrared spectrum) - mapping of about 15 top level biotopes (according to BKompV - Bundeskompensationsverordnung). Proof of technical viability of using 3D synthetic datasets (forest) to classify forests with the pre trained model (inference).
Jun-2023 Team BIO.KI.S.S. is winner of datarun2023 of BMDV for most technical approach for the project pitch for a 3d synthetic dataset pipeline for ML model training improvement. German Interview with Joerg Osarek Nov-2023 until Feb-2024 BIO.KI.S.S. GBT(Grüne Bahntechnik): small IT architecture sampling on a subset of tree species and forest biotopes. Proof of the datarun hypothesis that ML model training can be improved with highly accurate procedurally generated multi spectral synthetic 3d datasets. Core components of the 3DTWINZ TRAINBOW™ pipeline framework have been prototyped. We were able to improve the training of the ML model by adding high-quality synthetic 3D datasets and thus, for example, improve the recognition rate of birch trees in the TreeSat AI dataset from 46% to 57% - without further optimization. Participants: DB Netz AG / DB InfraGO, Fraunhofer IIS, 3dtwinz. Results show that training optimization can be improved further and also where to apply leverage for ML model optimization.Everyone is at a different point in their ML journey. You don't need to buy a pig in a poke. Choose the package that suits you best.
In a 45 minute remote call we can explore: Where are you today and where do you want to go with your ML pipeline and what are your top challenges where 3DTWINZ could be of help. You will receive an honest feedback whether or not we believe we can add value to your endeavour.
1 day personal remote or on site workshop for a small group. This one day workshop is about your personal ML journey. We cover:
Interested? Then feel invited to mail us at: contact at climatehackerz dot com or call Joerg Osarek: +49-151-23 0 24 333.