ImageNet CLIP Features: Precomputed Embeddings for Image Classification

A dataset of precomputed CLIP model features for ImageNet images, prepared by AbstractPhil. The dataset was last updated on October 2, 2025, and the preparer script is included. The splits are named train, val, and test for ease of use.

Use Cases

• Train image classifiers based on precomputed CLIP features.
• Benchmark transfer learning performance using the provided train/val/test splits.
• Analyze the feature space of the CLIP model on the ImageNet dataset.

Strengths

• Last updated on October 2, 2025, indicating recent maintenance.
• Includes the preparer script, allowing for reproducibility of the feature extraction process.
• Splits are explicitly named train, val, and test for standardized use.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

AbstractPhil on Hugging Face.

Collection Method

Features extracted from ImageNet images using a CLIP model with PyTorch configured for determinism.

Freshness

2025-10-02 16:52:22