lucataco/siglip 🖼️📝 → 📝

▶️ 2.1K runs 📅 Jan 2024 ⚙️ Cog 0.8.6 🔗 GitHub 📄 Paper ⚖️ License
image-embedding image-to-text

Performance

0.6sTypical run time
~72sCold start (first call)
2.1KTotal runs

About

SigLIP proposes to replace the loss function used in CLIP by a simple pairwise sigmoid loss

Example Output

Output

[{'score': 0.1979, 'label': '2 cats'}, {'score': 0.0, 'label': ' a remote'}, {'score': 0.0, 'label': ' a plane'}]

Performance Metrics

0.59s Prediction Time
72.29s Total Time
All Input Parameters 
{
  "image": "https://replicate.delivery/pbxt/KHHNwuUzEcvpu9PlzbqnyMnOLhkizlMqtXwOeYhzWCyKBhUi/cats.jpg",
  "candidate_labels": "2 cats, a plane, a remote"
}
Input Parameters
image (required) Type: string
Input image
candidate_labels Type: stringDefault: 2 cats, a plane, a remote
Candidate labels separated by commas
Output Schema

Output

Type: string

Version Details
Version ID
a39b979e4671e037ad90662ba6bd9377cbf0508083f2877b06e3f3850b8c8fa6
Version Created
January 23, 2024
Run on Replicate →