FavnaviImageEncoderV1 is a custom implementation, so trust_remote_code=True is required only when loading the model. AutoImageProcessor can be loaded without setting trust_remote_code=True.

python

import torch
from transformers import AutoModel, AutoImageProcessor
from transformers.image_utils import load_image

# Load the model and processor
model_name = "ly-corporation/favnavi-vision-ecommerce-v1-base"

model = AutoModel.from_pretrained(model_name, trust_remote_code=True).eval()
processor = AutoImageProcessor.from_pretrained(model_name)

# Load the image
image = load_image("https://huggingface.co/datasets/merve/coco/resolve/main/val2017/000000000285.jpg")
inputs = processor(images=[image], return_tensors="pt").to(model.device)

# Run inference
with torch.no_grad():
    image_embeddings = model.get_image_features(**inputs)

print(image_embeddings.shape)

python

from transformers import pipeline
from transformers.image_utils import load_image

pipe = pipeline(
    task="image-feature-extraction",
    model="ly-corporation/favnavi-vision-ecommerce-v1-base",
    trust_remote_code=True,
)

image = load_image("https://huggingface.co/datasets/merve/coco/resolve/main/val2017/000000000285.jpg")
embedding = pipe(image, return_tensors=True)
print(embedding.shape)

Models

Name	Pretrained model	Dim	In-house eval (mAP@20)	ILIAS small (mAP@1k)	ILIAS full (mAP@1k)
ly-corporation/favnavi-vision-ecommerce-v1-base	google/siglip2-base-patch16-naflex	256	0.861	31.9	27.7
ly-corporation/favnavi-vision-ecommerce-v1-large	google/siglip2-so400m-patch16-naflex	256	0.894	44.7	40.6

ILIAS small/full results are evaluated on the ILIAS dataset.
Training data: A subset of an in-house e-commerce dataset
Training objective: Trained with ArcFace over 535K product classes (the ArcFace class embeddings are not included in the model weights)

License

Apache-2.0

Citation

@misc{favnavi-vision-ecommerce-v1,
  title = {Favnavi Vision E-commerce V1: A Visual Feature Model for Product Retrieval in the E-commerce Domain},
  author = {Nishimura, Shuhei and Doi, Kenji and Yamashita, Fumiya and Yonebayashi, Dai and Iwasaki, Masajiro},
  year = {2026},
  howpublished = {\url{https://huggingface.co/ly-corporation/favnavi-vision-ecommerce-v1-base}}
}