in seed/util/preprocess.py [0:0]
def understand_image_with_gpt(client, deployment_name, image_path, caption="", max_tokens=1024, language="Korean"):
data_url = local_image_to_data_url(image_path)
if caption == "":
prompt = f"Describe this image in {language} language. "
else:
prompt = f"Describe this image in {language} language (note: it has image caption: {caption})."
response = client.chat.completions.create(
model=deployment_name,
messages=[
{ "role": "system", "content": "You are a helpful assistant." },
{ "role": "user", "content": [
{
"type": "text",
"text": prompt
},
{
"type": "image_url",
"image_url": {
"url": data_url
}
}
] }
],
max_tokens=max_tokens
)
img_description = response.choices[0].message.content
return img_description