Modified the official source code of OpenAI to enable clip not only to support PIL, but also to support opencv-python
import clip
import torch
import cv2 as cv
device = "cuda" if torch.cuda.is_available() else "cpu"
model, preprocess = clip.load("./ViT-B-32.pt", device=device)
image = preprocess(cv.imread("./images/dog.jpg")).unsqueeze(0).to(device)
text = clip.tokenize(["a diagram", "a dog", "a cat"]).to(device)
with torch.no_grad():
image_features = model.encode_image(image)
text_features = model.encode_text(text)
logits_per_image, logits_per_text = model(image, text)
probs = logits_per_image.softmax(dim=-1).cpu().numpy()
print(probs)