datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
openaccess-embeddings-siglip2
metmuseum/openaccess-embeddings-siglip2
Image embeddings for every public-domain artwork in metmuseum/openaccess, produced by google/siglip2-so400m-patch14-384.
Column
Type
Notes
objectID
int64
Primary key — matches objectID in metmuseum/openaccess
embedding
list<float32>
L2-normalised, dim = 1152
model
string
Source model id
dim
int32
Embedding dimension
Image bytes are not stored here; join against the main dataset to recover them.Embedding spec: dim=1152… See the full description on the dataset page: https://huggingface.co/datasets/metmuseum/openaccess-embeddings-siglip2.openaccess-embeddings-siglip2-naflex
metmuseum/openaccess-embeddings-siglip2-naflex
Image embeddings for every public-domain artwork in metmuseum/openaccess, produced by google/siglip2-so400m-patch16-naflex.
Column
Type
Notes
objectID
int64
Primary key — matches objectID in metmuseum/openaccess
embedding
list<float32>
L2-normalised, dim = 1152
model
string
Source model id
dim
int32
Embedding dimension
Image bytes are not stored here; join against the main dataset to recover them.Embedding spec:… See the full description on the dataset page: https://huggingface.co/datasets/metmuseum/openaccess-embeddings-siglip2-naflex.siglip2-large-lora-v1-dataset
SigLIP2-Large LoRA v1 — Training Pairs
Material/surface visual-similarity training pairs for fine-tuning
SigLIP2-large with LoRA + Supervised Contrastive loss. Each row links a
catalog product image to one polygon-clipped material crop from a SAM3-segmented
interior room render.
Statistics
Split
Rows
train
5,618
eval
576
Layout
.
├── dataset.jsonl train split (one JSON object per line)
├── eval.jsonl held-out eval split
└──… See the full description on the dataset page: https://huggingface.co/datasets/subhrokomol/siglip2-large-lora-v1-dataset.coco-2017-siglip2-embeddings
COCO 2017 SigLIP 2 Image Embeddings
Pre-computed image embeddings for the COCO 2017 dataset, generated with Google's SigLIP 2 (SoViT-400M, 384px).
Overview
Property
Value
Model
google/siglip2-so400m-patch14-384
Vector dimensions
1152
Normalization
L2-normalized (unit vectors)
Source dataset
COCO 2017
Image resolution
384 x 384 (resized by SigLIP 2 processor)
Dataset Structure
Schema
Each row contains the… See the full description on the dataset page: https://huggingface.co/datasets/jrmiller/coco-2017-siglip2-embeddings.siglip2-blind-spots
SigLIP2-Large Blind Spots Dataset
Model Tested
google/siglip2-large-patch16-512
878M parameters
Architecture: Vision Transformer with sigmoid contrastive loss
How The Model Was Loaded
Zero-Shot Classification
from transformers import pipeline
import torch
classifier = pipeline(
task="zero-shot-image-classification",
model="google/siglip2-large-patch16-512",
torch_dtype=torch.float16,
device=0
)
results = classifier(image… See the full description on the dataset page: https://huggingface.co/datasets/Dronezup/siglip2-blind-spots.siglip2-base-blindspotsaic2026-visual-siglip2-so400m