DataComp-VLM benchmark shows instruction-heavy data mixing outperforms filtering for VLM training, with DCVLM-Baseline achieving 63.6% on 33 tasks for 8B models (+5.4pp over FineVision).
arXiv preprint arXiv:2307.03132 , year=
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
fields
cs.CV 3years
2026 3representative citing papers
CPI ranks image-text pairs using phrase-level sensitivity scores from nonce substitutions to improve compositional performance in VL pretraining, achieving gains on relation benchmarks with a 50% data subset.
An iterative bootstrapped self-filtering approach selects balanced clean and diverse subsets from noisy vision-language datasets to train improved CLIP models.
citing papers explorer
-
DataComp-VLM: Improved Open Datasets for Vision-Language Models
DataComp-VLM benchmark shows instruction-heavy data mixing outperforms filtering for VLM training, with DCVLM-Baseline achieving 63.6% on 33 tasks for 8B models (+5.4pp over FineVision).
-
What Does the Caption Really Say? Counterfactual Phrase Intervention for Compositional Data Selection in Vision-Language Pretraining
CPI ranks image-text pairs using phrase-level sensitivity scores from nonce substitutions to improve compositional performance in VL pretraining, achieving gains on relation benchmarks with a 50% data subset.
-
Data Selection Through Iterative Self-Filtering for Vision-Language Settings
An iterative bootstrapped self-filtering approach selects balanced clean and diverse subsets from noisy vision-language datasets to train improved CLIP models.