Steel scrap composition predicted from sparse handheld pile images
SteelHuman: Predicting Operator Estimates of Steel-Scrap Composition from Sparse RGB Views
Computer Vision and Pattern Recognition
Summary
Estimating what types of steel scrap are in a large pile is important but usually slow and requires expert workers. The authors study a simpler way to guess the pile's composition by using just a few photos taken by hand from different angles. They use a special image feature extractor, frozen DINOv2 tokens, to combine information from these pictures and get estimates close to those made by human operators. Their method also performs well on a publicly available steel classification data set, outperforming previous approaches.
What this means in practice
- •For steel mill operators: Estimate steel scrap composition quickly from sparse images to aid yield planning and supplier compensation.
- •For industrial quality inspectors: Use handheld image capture to support fast assessment of material composition in large scrap piles.
Authors
Jovan Kamchev, Zoran Ivanovski
Abstract
Steel scrap composition estimation is an important part of yield planning and fair supplier compensation. Measuring the composition of a multi-ton pile is an inherently laborious process, so the assessment is often left to trained human operators. Recent works tackle composition-level tasks, but require structured acquisition and multi-stage preprocessing. We study a simpler approach to steel scrap load assessment by aggregating features from sparse, handheld views of the pile. On 600 out-of-fold piles, frozen DINOv2 tokens aggregated uniformly across views reach 13.4% total variation (TV) distance against operator estimates. Most remaining error lies in allocation among four thickness-defined heavy melting scrap (HMS) classes, yet image-dependent allocation reduces within-HMS TV from 25.1% under a training-set split prior to 11.2%. Due to the lack of public datasets and benchmarks for predicting composition, external evaluation relies on the public DOES dataset intended for classification. A classifier using the same frozen features reaches 93.97% accuracy, exceeding the published results we compare against. The composition model also aligns with related DOES categories without adaptation.