What image/video training data is hardest to find right now? [R]

An AI developer is crowdsourcing the most critical missing piece of the modern AI stack: high-quality, specific training data. Through a viral post, they are asking the machine learning community to identify the image and video datasets they desperately need but cannot find, aiming to build a platform that systematically fills these gaps. The proposed system would leverage contributors' smartphones to capture real-world scenes, automatically annotate them using models like YOLO for object detection and CLIP for understanding, and enrich each image with a robust suite of over 40 metadata fields including GPS location, weather conditions, timestamps, and OCR-extracted text.

This initiative highlights a fundamental bottleneck in AI development: while model architectures advance rapidly, progress is often gated by the availability of clean, well-labeled, and context-rich data for specific domains. The founder's suggested categories—such as European street scenes (notably lacking for Switzerland and France), supermarket shelves with OCR-extracted prices for dynamic pricing analysis, and analog utility meters for automation projects—point to real-world business and research applications currently hindered by data scarcity. The community's response will directly shape the platform's first collection targets, creating a demand-driven pipeline for the next generation of computer vision models.