Semasia
Collection
Large-scale dataset of latent embeddings from ~1700 pretrained vision models across 8 image benchmarks. Enables studying latent geometry, benchmarking • 9 items • Updated
The dataset viewer should be available soon. Please retry later.
Latents for celeba (timm)
This repository hosts precomputed embeddings for celeba across many timm models.
Each config corresponds to a single model;
only that model's Parquet files are read on load_dataset.
Usage
from datasets import load_dataset
ds_test = load_dataset("spaicom-lab/semasia-celeba", "aimv2_1b_patch14_224.apple_pt", split="test")
ds_train = load_dataset("spaicom-lab/semasia-celeba", "aimv2_1b_patch14_224.apple_pt", split="train")
ds_valid = load_dataset("spaicom-lab/semasia-celeba", "aimv2_1b_patch14_224.apple_pt", split="valid")
Notes
- Configs are generated from what is actually uploaded on the Hub (parquet presence).
- Based on flwrlabs/celeba
- Downloads last month
- 11,572