A standalone ONNX vision projector extracted from the FastVLM pipeline.
Converts image features into the FastVLM multimodal embedding space and is
designed for CPU-based image-to-text and vision+language inference.
Use this model to convert images into embeddings for a compatible FastVLM
language backbone.
1import onnxruntime as ort
2
3session = ort.InferenceSession("vision_projector_v1_standalone.onnx")
4# feed in your preprocessed image tensor and run inference
