OpenCLIP with OpenVINO

Original OpenCLIP repo and models.

To convert model

pip install -e git+https://github.com/mlfoundations/open_clip.git#egg=open_clip_torch
python convert_clip.py ViT-B-32 -c <path/to/ckpt/file.pth> -o models/

To run the model

python clip.py -v models/vit_b_32_visual.xml -t models/vit_b_32_text.xml -i <path/to/images/cat_dog.jpeg> -p "eagle,cat,tiger,dinosaurs" -d 'GPU'

Run benchmark

benchmark_app -m models/vit_l_14_visual.xml -hint latency -d GPU -data_shape "x[1,3,224,224]"
benchmark_app -m models/vit_l_14_text.xml -hint latency -d GPU -data_shape "text[1,77]"