Adding nearest-neighbor and cross nearest-neighbor supervision from frozen pretrained unimodal encoders to the CLIP loss improves lightweight vision-language models on zero-shot and retrieval benchmarks.
An inverse scaling law for clip training,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
CLIP-PING: Boosting Lightweight Vision-Language Models with Proximus Intrinsic Neighbors Guidance
Adding nearest-neighbor and cross nearest-neighbor supervision from frozen pretrained unimodal encoders to the CLIP loss improves lightweight vision-language models on zero-shot and retrieval benchmarks.