Tag: #dataset24 results
A user shares a 150k+ stool images dataset for training computer vision models, detailing a manual verification process for annotations like Bristol type, consistency, and blood indicators. They train iteratively on verified data and seek advice on scalable workflows. The post asks for ML best practices in high-stakes annotation domains.
A security firm has released a massive dataset comprising 114 million records derived from live enterprise attack traffic. The data was captured across five enterprise networks in 2024, offering real-world insights for threat analysis.
nohurry urges users to stop using his filtered version of the Opus-4.6-Reasoning-3000x dataset, as Crownelius has updated the original. The filtered set was a temporary fix for refusals. Donate to Crownelius via ko-fi.
NY figurative artist publishes 3-4k images spanning five decades on Hugging Face, focusing on human figure evolution. Includes structured metadata, various media, CC-BY-NC-4.0 license. Useful for style evolution, representation learning, ethical training data.