A learned product-quantization VAE with a shared codebook compresses CLIP image features to about 2-3 x 10^-3 bits per pixel with little loss in downstream semantic task accuracy.
Variational image compres- sion with a scale hyperprior,
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
-
Compression Beyond Pixels: Semantic Compression with Multimodal Foundation Models
A learned product-quantization VAE with a shared codebook compresses CLIP image features to about 2-3 x 10^-3 bits per pixel with little loss in downstream semantic task accuracy.