TIPSv2 — B/14
TIPSv2 (Text-Image Pre-training with Spatial awareness) is a family of contrastive vision-language models that produce spatially rich image features aligned with text embeddings. This is the Base variant with 86M vision params and 110M text params. Try the code snippets below or check out the GitHub repo for more use cases and visualizations, including zero-shot segmentation.
| Variant | Vision params | Text params | Embed dim | DPT Heads |
|---|---|---|---|---|
| B/14 | 86M | 110M | 768 | B/14-dpt |
| L/14 | 303M | 184M | 1024 | L/14-dpt |
| SO400m/14 | 412M | 448M | 1152 | SO400m/14-dpt |
| g/14 | 1.1B | 389M | 1536 | g/14-dpt |
Usage
pip install transformers torch torchvision sentencepiece scikit-learn
Load the model
from transformers import AutoModel
model = AutoModel.from_pretrained("google/tipsv2-b14", trust_remote_code=True)
model.eval()
Encode images
Images should be tensors in [0, 1] range (just ToTensor(), no ImageNet normalization).
from torchvision import transforms
from PIL import Image
import requests
transform = transforms.Compose([
transforms.Resize((448, 448)),
transforms.ToTensor(),
])
url = "https://huggingface.co/spaces/google/TIPSv2/resolve/main/examples/zeroseg/pascal_context_00049_image.png"
image = Image.open(requests.get(url, stream=True).raw)
pixel_values = transform(image).unsqueeze(0)
out = model.encode_image(pixel_values)
print(out.cls_token.shape) # (1, 1, 768) — global image embedding
print(out.patch_tokens.shape) # (1, 1024, 768) — per-patch spatial features
Encode text
text_emb = model.encode_text(["a photo of a bus", "a photo of a dog"])
print(text_emb.shape) # (2, 768) — one embedding per query
Zero-shot classification
import torch.nn.functional as F
classes = ["bus", "car", "dog", "cat"]
cls = F.normalize(out.cls_token[:, 0, :], dim=-1)
text_emb = F.normalize(model.encode_text(classes), dim=-1)
similarity = cls @ text_emb.T
print(classes[similarity.argmax()]) # bus — predicted class
Visualize spatial features
import numpy as np
from sklearn.decomposition import PCA
spatial = out.patch_tokens.reshape(1, 32, 32, 768)
feat = spatial[0].detach().cpu().numpy().reshape(-1, 768)
rgb = PCA(n_components=3, whiten=True).fit_transform(feat).reshape(32, 32, 3)
rgb = 1 / (1 + np.exp(-2.0 * rgb)) # sigmoid for [0, 1] range with good contrast
print(rgb.shape) # (32, 32, 3) — PCA of patch features as RGB
GPU inference
model = model.cuda()
out = model.encode_image(pixel_values.cuda())
text_emb = model.encode_text(["a city"])
Model details
- Architecture: ViT vision encoder (12 layers) + Transformer text encoder (12 layers)
- Image preprocessing: resize to any resolution, convert to
[0, 1](no ImageNet normalization) - Text preprocessing: SentencePiece tokenizer, lowercased, max 64 tokens
- Patch size: 14x14 pixels
License
Apache 2.0
Citation
@inproceedings{cao2026tipsv2,
title = {{TIPSv2: Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment}},
author = {Cao, Bingyi and Chen, Koert and Maninis, Kevis-Kokitsi and Chen, Kaifeng and Karpur, Arjun and Xia, Ye and Dua, Sahil and Dabral, Tanmaya and Han, Guangxing and Han, Bohyung and Ainslie, Joshua and Bewley, Alex and Jacob, Mithun and Wagner, Rene and Ramos, Washington and Choromanski, Krzysztof and Seyedhosseini, Mojtaba and Zhou, Howard and Araujo, Andre},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2026},
url = {https://arxiv.org/abs/2604.12012}
}