README.md

---
dataset_info:
  features:
  - name: image_id
    dtype: string
  - name: label
    dtype: int32
  - name: clip_model
    dtype: string
  - name: clip_features
    list: float32
  - name: vector_dim
    dtype: int32
  - name: timestamp
    dtype: timestamp[ns]
  splits:
  - name: clip_vit_b32_train
    num_bytes: 2723761042
    num_examples: 1281167
  - name: clip_vit_laion_b32_train
    num_bytes: 2789100559
    num_examples: 1281167
  - name: clip_vit_laion_b32_validation
    num_bytes: 108850000
    num_examples: 50000
  - name: clip_vit_b16_train
    num_bytes: 2777570056
    num_examples: 1281167
  - name: clip_vit_b16_validation
    num_bytes: 108400000
    num_examples: 50000
  - name: clip_vit_l14_train
    num_bytes: 4090766231
    num_examples: 1281167
  - name: clip_vit_l14_validation
    num_bytes: 159650000
    num_examples: 50000
  - name: clip_vit_laion_bigg14_train
    num_bytes: 6728689084
    num_examples: 1281167
  - name: clip_vit_laion_bigg14_validation
    num_bytes: 262600000
    num_examples: 50000
  - name: clip_vit_b32_validation
    num_bytes: 108400000
    num_examples: 50000
  - name: clip_vit_b32_test
    num_bytes: 216800000
    num_examples: 100000
  - name: clip_vit_b16_test
    num_bytes: 216800000
    num_examples: 100000
  - name: clip_vit_laion_b32_test
    num_bytes: 217700000
    num_examples: 100000
  - name: clip_vit_l14_test
    num_bytes: 319300000
    num_examples: 100000
  - name: clip_vit_laion_h14_test
    num_bytes: 422500000
    num_examples: 100000
  download_size: 25438949728
  dataset_size: 21250886972
configs:
- config_name: default
  data_files:
  - split: clip_vit_b32_train
    path: data/clip_vit_b32_train-*
  - split: clip_vit_b32_validation
    path: data/clip_vit_b32_validation-*
  - split: clip_vit_laion_b32_train
    path: data/clip_vit_laion_b32_train-*
  - split: clip_vit_laion_b32_validation
    path: data/clip_vit_laion_b32_validation-*
  - split: clip_vit_b16_train
    path: data/clip_vit_b16_train-*
  - split: clip_vit_b16_validation
    path: data/clip_vit_b16_validation-*
  - split: clip_vit_l14_train
    path: data/clip_vit_l14_train-*
  - split: clip_vit_l14_validation
    path: data/clip_vit_l14_validation-*
  - split: clip_vit_laion_bigg14_train
    path: data/clip_vit_laion_bigg14_train-*
  - split: clip_vit_laion_bigg14_validation
    path: data/clip_vit_laion_bigg14_validation-*
  - split: clip_vit_b32_test
    path: data/clip_vit_b32_test-*
  - split: clip_vit_b16_test
    path: data/clip_vit_b16_test-*
  - split: clip_vit_laion_b32_test
    path: data/clip_vit_laion_b32_test-*
  - split: clip_vit_l14_test
    path: data/clip_vit_l14_test-*
  - split: clip_vit_laion_h14_test
    path: data/clip_vit_laion_h14_test-*
task_categories:
- feature-extraction
- image-feature-extraction
license: mit
tags:
- features
- image_features
- extracted_features
- precomputed_features
- imagenet
- imagenet_features
- clip_vit
- variants
size_categories:
- 1M<n<10M
---

# Update: 10/2/2025

Claude said that I'm not being careful enough with my database curation after grilling me for 20 minutes, so I included the preparer script as well.

Claude Sonnet 4.5 is kind of a chad.

# Update; 9/26/2025

Having to download this whole repo is annoying, so I'm making sure the splits are named train/val/test (if they exist) and the named subset is the clip name.


# Older non-dated updates

Everything extracted with torch configured as deterministic; using seed 42 on an a100 using colab; so if it has variances from expectation it's on cuda.

It's a little quirky; 
* Most of the splits have train, test, val. Many do not.
* Most of the splits have a proper "image_id" md5 id for verification.

The prompts used were direct literal prompts for the classification name;

No use of "a photo of" or any such invariance; just the classification text.

This is a series of clip-vit extracted feature maps from a 256x256 cropped and resized imagenet variant hosted here on huggingface.

I ran the processor 224x224 and then extracted features from the entire dataset batch-sequentially while simultaneously capturing the necessary classifiers and classifications associated with the images for downstream testing and assessment.

Academic and research purpose use only.

clip-vit-large-patch14 variations do exist in the splits. 

clip-vit-bigG is the 1280 dim variation and it does exist; it took quite a while to extract - and it is in fact missing it's test split. Sorry about that.

There are many variants of clip-vit-base from many variant forms. Each of them extracted using the same process as the others.