Classification sourcing begins with the decision your model must make from a photograph. Share the categories, intended environment, and examples of mistakes that matter. Photo Datasets is a Fiund inquiry point for reviewing that brief and exploring potential sources; it is not a catalog of approved stock.
Make labels operational
Explain whether each image receives one class, multiple classes, or attributes attached to individual objects. Include accepted examples, neighboring categories, and an uncertainty policy. Source descriptions and existing folder names may not match your task. A pilot should establish whether the proposed material can support consistent labels without asking reviewers to infer information that is not visible.
Acquire meaningful variation
Describe relevant differences in photographed subjects, locations, viewpoints, and lighting. Request both image counts and source-group counts where available. Repeated images can be useful but should remain identifiable. Define what must be unfamiliar during evaluation and preserve the relationships needed to keep related examples in the appropriate split rather than relying on a random allocation alone. Keep class names, numeric identifiers, and any label mappings in the same release. This allows future sourcing rounds to expand coverage without silently changing what an existing target means.
Review before committing to scale
Ask for representative files, the intended annotation export, and a draft manifest. Inspect difficult classes separately from ordinary examples, then document accepted properties and unresolved gaps. Discuss source documentation, proposed commercial uses, sample availability, and delivery scope for the specific project. Email Jaeden with the requirements below so the sourcing conversation starts with an actionable brief.
Include in your inquiry
- Target categories, attributes, and uncertainty rules.
- Expected image conditions and independent coverage goals.
- Grouping and evaluation separation requirements.
- Intended use, source documentation needs, and timeline.