A contrastive training loss that groups the image regions most similar to a referring expression, using the true object count, improves counting accuracy by over 22% on REC-8K.
Amini-Naieni, T
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving Contrastive Learning for Referring Expression Counting
A contrastive training loss that groups the image regions most similar to a referring expression, using the true object count, improves counting accuracy by over 22% on REC-8K.