metadata
library_name: transformers
tags:
- vehicle
- metric-learning
- image-embedding
pipeline_tag: image-feature-extraction
Vehicle Cosine Encoder
256-D embedding model for vehicle re-identification. Trained with CosineEmbeddingLoss on ~5k vehicles from used-car marketplace listings.
Usage
from transformers import AutoModel
model = AutoModel.from_pretrained("quebeccyb/<repo-name>", trust_remote_code=True).eval()
# pixel_values: float tensor (B, 3, 256, 256), letterboxed to square
emb = model(pixel_values)