Catalogue of Services

This service concerns training AI models on behalf of the customer for a specific task and optimisation objective, e.g., improving accuracy on crop classification from image data. The target model is the solution provided by the customer that needs to be enhanced with respect to a set of pre-determined features to reach the desired performance level. However, if required, the training can also be applied to additional state-of-the-art models available in the market for benchmarking purposes. If not defined by the customer, some features of the training process can be identified via service S00179 (desk assessment activities for digital systems and/or data): for instance, model features to improve, reference model baselines to include in the performance comparison, as well as benchmark datasets. The data used for training the model can be either provided by the customer or annotated ad hoc as a preparatory activity to model training (via service S00290 - Data Labelling); another possibility is that data are retrieved among reference benchmark datasets that are openly available. We will also agree with customers on the level of hardware acceleration required, based on the considered AI models: e.g., GPU acceleration via connection to a remote server vs. on-device training.

The tests in this service are aimed at products that run complex apps (based on a multitude of microservices, including heavy AI algorithms) in the fields, with limited to no connectivity to the cloud. In these cases, we propose a benchmarking designed to understand what the bottlenecks are in the implemented solution architecture to help TEF customers improve the efficiency and reliability of such products/solutions. The service may also encompass virtual facility elements (enriched datasets, virtual machines hosting containerised modules) blended with physical sensors and infrastructure in our test field or directly installed at the client's premises if need be for the duration of the experimentation.

Through this service, we provide specialised technical systems (both hardware and software) and expertise for collecting relevant data during testing.

The scope of the data collected can vary, encompassing information generated by the agrifoodTEF technical infrastructure (e.g., through the installation of IoT sensors), data from the system under test, and ground truth data for system evaluation. If needed, this service can facilitate the interconnection between agrifoodTEF's data collection systems and the customer's systems.

In addition to the raw data collected, we will supply customers with comprehensive documentation detailing the logged features, conditions of the testing environment at the time of testing, and any parameter values, variation ranges, and specifics necessary for reproducibility.

When required, data collection and documentation activities may involve field experts such as engineers and agronomists.

Please note that annotation and labeling of collected data, if necessary, can be provided through service S00115. 

One of the key activities during digital testing is the collection of data concerning the progression and the final outcome of the tests. Such data enable the evaluation of system performance by the customer or – if needed – by AgrifoodTEF (via Service S00184). This service manages the collection of data relevant to performance evaluation produced during the tests by both the system under test and the computational environment where the tests take place. 

Examples of collected data comprise information produced within a virtual environment to simulate sensor data collection in a physical environment; statistics about AI model performance in the test and deployment phase (e.g., occupied memory, number of trainable parameters, training/optimisation loss, etc.); specific labels and annotations to use as ground truth for evaluating the system; and system output when subjected to a range of test conditions. The minimum set of data to be collected is defined by the evaluation metrics that the user chose (either on their own or with AgrifoodTEF support, via Service S00178) to process them; generally, a larger set of data wrt the minimum is selected by AgrifoodTEF together with the customer to provide a richer view of the system’s performance and to enable the application of other metrics in the future, if needed. As an output of the service, in addition to the raw data, we also provide the customer with documentation describing logged features and conditions of the testing environment at the time of testing, as well as any parameter values, variation ranges and specifics required for reproducibility purposes.

This service consists of deploying a modular sensor payload (equipped with Lidar, inertial motion units, GNSS, Radar, stereo cameras, and ultra-wideband technology) to agricultural environments such as vineyards and orchards to collect and annotate high-fidelity datasets tailored to customer needs. Data are systematically labelled and can be delivered to agrifoodTEF customers as structured datasets for AI-driven agricultural solutions.

This service concerns the curation of ground truth annotations for testing and experimentation data. More specifically, data labelling concerns the association of portions of data (e.g., image regions) to descriptions that provide information about the significance of those portions. In particular, labelling is required whenever the data are meant to be used for model training. We also provide labelling services for domain-specific data attributes, such as disease indicators and their confounding factors, by experts in agronomy. The labelled data will be accompanied by a report synthesising quality metrics and statistics like the number of data points annotated, the percentage of the full set covered for the annotations, the number of annotators and the inter-annotator agreement.

Through this service, we offer customers the ability to validate and augment the data collected during the different testing phases. Examples of operations that this service can provide include quality checks (e.g., presence of all datastreams, presence of drop-outs, technical quality such as focus or exposure of camera images, agronomic significance of data); preparation for performance evaluation (e.g., identification of the images containing selected visual markers or specific plants, association between images from the system under test and the ground truth); data processing (e.g., transcoding) and augmentation (e.g., generation and incorporation of metadata). Data labelling (e.g., to enable training of AI models) is not included, as it is the object of a separate service (S00291).

Test activities involve three key components: the environment (where the tests are conducted), the protocol (defining what tests are performed and how), and evaluation metrics (used to assess the test results). Through this service, we assist customers in designing a computational environment tailored to running the digital tests required for their software components. Specifically, this service:

• Defines the data (and metadata) requirements needed to support the test
• Identifies the necessary software environment (e.g., operating system, libraries)
• Selects an appropriate simulator and defines simulated environments, if required
• Chooses remote access tools, if applicable.

As a result, we deliver a comprehensive design of the testing environment, including specifications such as memory requirements, GPU/TPU acceleration, and necessary libraries. This environment can be deployed on either the agrifoodTEF digital infrastructure or the customer's computing infrastructure (i.e., on premises).

Any test activity involves three main components, i.e., environment (where the tests take place), protocol (defining what activities are executed and how), and evaluation metrics (used to assess the results of the tests). This service concerns the last element; its goal is to design the best metrics to evaluate the performance of digital systems such as (for instance) AI models or computer vision software. The digital environment and the testing protocol metrics can be designed—if required—via services S00176 and S00177. Our team will identify and define with customers the most adequate set of quantitative metrics to assess the outcome of the digital testing activities. In order to ensure the relevance of the metrics with respect to the real-world use cases, the team will involve engineers and agronomists. The metrics will be adapted not only to the task that the digital system under test (e.g., a piece of software) is designed to perform, but also to the features of the data used for the tests. For instance, a customer that has developed a machine incorporating an AI model will be interested in testing the model on data generated by their own machine: the performance metrics will therefore need to be adapted to the specific features of that data.

Any physical test activity involves three main components: environment (where the tests take place), protocol (defining what tests are executed and how) and evaluation metrics (used to assess the results of the tests). This service concerns the last element; its goal is to design the best metrics to evaluate the performance of a customer solution taking into consideration the use cases specified by the customer and the environment and protocol chosen for the tests (which, if needed, can be designed via services S00106 and S00107). Our team will identify and define with customers the most adequate set of quantitative (i.e., based on instrumental measurements) and/or qualitative (i.e., relying on expert human judgement) metrics to assess the system functionalities of interest. This phase will involve, in particular, agronomists and experts in agricultural machinery. Based on the defined evaluation metrics, a set of requirements for the collection of required data and ground truth annotations will also be defined accordingly. For instance, the service may lay out the specifications for dedicated data collection campaigns (possibly executed via service S00113). This phase will involve engineers and experts in AI and robotics. On request, the output of the service will include analyses on additional environmental factors than those directly tracked through the designed metrics (e.g., seasonal effects, impact of test distribution over time on results).

Any physical test activity involves three main components: environment (where the tests take place), protocol (defining what tests are executed and how) and evaluation metrics (used to assess the results of the tests). This service concerns the first element, i.e., the design of a physical testing environment for use cases such as (for instance) weeding, plant phenotyping, and precision spraying solutions. The protocol and the evaluation metrics can – if required – be designed via services S00107 and S00108. Depending on your requirements and reference system/solution to be tested, our team will design an ad hoc setup equipped with all the required features for testing. Environmental features include, for example, the crop and weed species to be prepared and their growth stage. - the plant layout and intra-row configuration, - seasonal weather and climate-related conditions (e.g., lighting conditions, wind, rain), - the type of soil, moisture level, and terrain conditions (e.g., uneven terrain, presence of any slopes, and so forth), - the technical infrastructure supporting the tests (e.g., electrical layout, network infrastructure, environmental sensors, data acquisition systems…). In order to consider all aspects of the environment, this service involves a team comprising both engineers and agronomists.

Any physical testing activity comprises three main components:

• the environment (where the tests are conducted),
• the protocol (which defines the tests to be executed and their methodology),
• and the evaluation metrics (used to assess the test results).

This service focuses on the protocol, while the environment and metrics can be designed as needed through services S00106 and S00108.

In the context of testing customer solutions within physical facilities, this service aims to create a suitable testing protocol based on the use cases specified by the customer. The components of the testing protocol defined in this phase may include:

• Defining the different phases of the protocol and their duration
• Outlining the operations to be executed in each phase
• Quantifying the effort required for each phase, including the number and qualifications of personnel involved
• Distributing testing operations over time, considering seasonal and daily variations, as well as weather conditions
• Establishing acceptable and desired variation ranges for each configurable element in the testing environment
• Determining the number of test repetitions required and their distribution across the variation ranges

To ensure a comprehensive approach to protocol design, this service involves a collaborative team of both engineers and agronomists.