Visual Intelligence and 3D-Modelling for next-level Retail
Collaborate with a leading multinational retail-group and leverage new technological advancements in visual intelligence to automate 3D-modelling. Transform the way retailers handle product imagery to enable faster and more accurate inventory management, improved customer experiences, and optimized supply chains.
For the roll-out of a scalable solution, there would be an adequate multi-million project budget available after the challenge.
#RetailInnovation #VisualIntelligence #3DModelling
✅ Completed 🏁 Winner Congratulations to VRee AI & Trillion Technology!
🏆 Prize EUR 5,000 prize pool + potential collaboration and an adequate multi-million project budget available
🪙 AI & Data Science This challenge is part of the AI & Data Science Innovation Program
In the following you will find all information, requirements and inspirational content to build relevant solutions. Always keep in mind that a final solution does not have to be in place at the first submission, but that the project intentionally follows a multi-step process.
Test images and relevant examples
Example articles can be found in typical retail stores! Besides easy product shapes - like tetra packs or simple boxes - modelling of more complex packages is required (in the long-run). The following are known challenges that you could look at:
- Glasses with honey – difficult color shading
- Bottles containing liquids, the more colored (including both bottle surface and the contained liquid) the more significant
- Red wine bottles including light reflections
- Plastic-wrapped cooled/frozen goods like pieces of salmon
MVP scenarios for inspiration
The long-term solution needs to be architected in holistic end-to-end use cases. To get there, you could already work on MVP scenarios like the following:
- Virtual mini-store, e.g. a shelf containing 10 products of different type and complexity. Via a simple user interface products can be picked and virtually rotated. Article master data and visually extracted information are displayed additionally on user request
- Combination of articles into larger units e.g. packaging of milk cartons, vegetables in glasses, water in plastic bottles, … virtual placement of multiple units into shelves or mixed pallets
Criteria to follow
- The 3D-modelling should be based on object photos
- Final solution should be able to process photos from various sources (smartphone, professional photo boxes, etc.) and should handle several types of packaging (cardboard, glass, plastic, etc.)
- The solutions should be implemented in modular patterns e.g. as individual capability services that can be orchestrated. Monolithic black-box architectures are not favored.
- Ultimately, the solution should be able to provide different "quality level 3D-models" based on the given need (such as highly photorealistic models and highly performant models with both low and high polygon count).
- Proprietary procedures and fee-based services should be avoided as much as possible.
Current efforts
The client is currently using industrial cameras and other sensors to scan articles in terms of 360°-photos, volume and weight. In a prototypical approach textured 3D-models using photogrammetry (for easy products) or manual modelling (for complex products) are generated.
An end-to-end solution (see MVP scenarios) has not yet been implemented.