A CLIP-based pipeline for cleaning noisy multi-label manufacturing image data, tested on Factorynet, reduces the label vocabulary from 6,426 to 408 distinct labels through similarity scoring and clustering.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Sanitizing Manufacturing Dataset Labels Using Vision-Language Models
A CLIP-based pipeline for cleaning noisy multi-label manufacturing image data, tested on Factorynet, reduces the label vocabulary from 6,426 to 408 distinct labels through similarity scoring and clustering.