PyTorch Image Captioning BLIP + CLIP Streaming CPU

Model: PyTorch Image Captioning — BLIP (candidate generation) + CLIP (ranking) Accelerator: CPU only Host: 50 × n1-standard-4 (4 vCPUs, 15 GB RAM)

This streaming pipeline performs image captioning using a multi-model open-source PyTorch approach. It first generates multiple candidate captions per image using a BLIP model, then ranks these candidates with a CLIP model based on image-text similarity.

The following graphs show various metrics when running PyTorch Image Captioning BLIP + CLIP Streaming CPU pipeline. See the glossary for definitions.

Full pipeline implementation is available here.