SigLIP-2

Google DeepMind · 2025-02 · image, text · discriminative · anchor

Successor to SigLIP, adding captioning-based pretraining and self-distillation objectives to the sigmoid loss.

Variants

SigLIP-2 ViT-B/32, SigLIP-2 ViT-B/16, SigLIP-2 ViT-L/16

Findings

• FX-002FIXLIP gives SigLIP-2 higher pointing-game recognition than CLIP at ViT-B/32 and ViT-B/16

Shared mechanisms

• Explanation faithfulnessalso in BakLLaVA, CF2, Claude 3, Claude 3.5, CLIP / CLIP-ViT (LC), DRUM, Fuyu, GEM, Gemini 1.5 / Gemini Pro 1.5, Gemma 2, GPT-3.5 / ChatGPT-3.5, GPT-4 / ChatGPT4 / GPT-4 Code Interpreter / GPT-4 Technical Report, GPT-4o, GPT-J, Idefics, Llama 3.1, LLaVA-NeXT / LLaVA 1.6, MAE-B/16, MobileNetV2, mPLUG-Owl3, OpenFlamingo, PGExplainer, Pythia, RCExplainer, ResNet / ResNet-152 / ResNet-101 / ResNet-50-BN, SigLIP, Stable Diffusion, TAGExplainer, ViT