An Innsbruck based museum, contact us to optimize the distribution of 3,299 paintings into their new storage facilities outside the city. The storage space is composed by an array of double side movable metal grid walls. The museum wants to optimize the painting arrange distribution on the walls to use the grid in the most efficient way.
The museum also established different classifications and considerations to hang the paintings on the wall:
- Use the minimal possible space required to hang the paintings.
- Some paintings needs to hang together in the same grid wall. For example, a series of paintings from the same artist, cronology order and/or object type.
- Paintings larger than X,Y dimensions or heavier than X Kg should be hang in the bottom of the grid wall for easy manipulation.
- Painting with a greater depth than Z dimension should be hang out.
- Reserve areas (consider aspect ratio) for paintings in exhibitions, restorations, etc.
- The distance between paintings in the wall should be related to their dimensions. Bigger objects require more space to manipulate than small ones.
- Another special requirements.
The database received was composed by a series of excel workbooks and folders with the painting pictures. The client need a report for all grid walls, showing the correct painting arrangement, including the relative location
(X, Y) on the wall and the right painting code ID. The graphic information displayed in the report should also be reflected in a table indicating wall number and the complete painting data: ID, Artist, Region, Year, Description, Dimension, Location, etc.
In the work process several software were written to achieve efficiently the objectives in time and budget. The workflow could be divided in 3 different stages: Pre-processing information, which consisted in analyze the existent data base. In the Optimization stage the software developed was designed to take intelligent decisions. Finally, the Display stage, consisted in generate the plan layouts and the different tables to add in the report. Although the strategy is presented trough individual stages, in reality the whole workflow run continuous. In some cases the workflow also ran back and forward to incorporate in the process new parameters.
In this stage, the database was analyzed to plan an strategy to make it useful for the classification and optimization process. Different codes were written (C# and VB in Excel) to survey and explore the whole database to detect voids and errors in the information. For example, some database entries related to the painting dimensions were swapped between Width and Height. During the pre-processing stage an image software processing was developed to resize and scale all painting pictures because many of them were big files and it was important also, to control from the beginning the outcome file sizes.
We worked close with the client in a workshop base format. Several working sessions were organized until the database was consistent and robust to be used in the optimization process.
Pre-Processing data stage. On the right a database example received from the client and on the left, the algorithm classify the errors and void information with different colors.
A decision tree algorithm was in charge to distribute the paintings on the grid walls. Nevertheless, some optimization strategies were implemented through the laga library to help the main algorithm.
A Decision Tree is a popular algorithm in the Artificial Intelligence field. In Machine Learning is used extensively for supervised learning classification problem, which map an input to a discrete label, like True or False. Given an input picture the model return True if is a Male, False is a Female. Decision trees are useful to analyse both numerical and categorical data. Since decision trees are not based on equations, are non-parametric models. In contrast, they are based on a simple recursive structure that work very fast.
The design of a decision tree algorithm is the result of the pre-processed training data
Decision Tree representation: The algorithm build it's structure based in the input data. The objective is narrow down the space search to find the best question and the write answer
The Algorithm builds in a particular order a number of questions to narrow down the space search, in order to get as much information out from that question towards to figure out where to locate the painting on the grid wall. The objective is always to reduce the number of possibilities, because it implies that the usefulness of a question depends upon the answer in the previous question.
Decision Tree Algorithm:
- Pick the best attribute, ideally split the data (grid wall) in halfs.
- Formulate a question
- Follow the answer path
- Locate the painting in the right position
- Repeat from step 1.
Optimal painting distribution examples
The information generated though the algorithms was compiled in a report and distributed to the team: The company in charge to hang the paintings on the walls and the museum staff.
The report and one of the tables with the painting locations. On the right image, the report assigned to the grid walls.
The company in charge to transport and hang the paintings on the grid walls finished in less time than expected, according to the work plan. And no errors were found in the optimization of the 3,299 paintings. The client was happy, and some museum staff members qualify the work done as a benchmark to look in the future for similar works. They also told us that staff members from other museums have been visit the storage to understand the process developed.
Example of one the grid walls
OneCanvas.io is a startup created in the information sharing age to help museums and art galleries to make the most of their database using advanced algorithms. Our expertise blends AI, data visualization and architectural design experience to bring the needed information to the palm of users.
Intelligent data services
- 1C provides Intelligent data services for the art world.
- 1C creates the bridge-link between the database, the art experience and the collection management.
- 1C unlocks and explores inherent potentials of a database.
- 1C fills the gap between art professionals and IT departments.