bdsqlsz commited on
Commit
d321083
1 Parent(s): 9a8ef70

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +1 -1
app.py CHANGED
@@ -10,7 +10,7 @@ def create_captions_llava_llama3_docci(image):
10
  pipe = pipeline('Lin-Chen/open-llava-next-llama3-8b')
11
  gen_config = GenerationConfig(repetition_penalty=1.10)
12
  image = Image.fromarray(np.uint8(image)).convert('RGB')
13
- response = pipe(('As an AI image annotation expert, provide accurate captions for this images to enhance the T5 model understanding of the content. Prioritize captions based on relevance. Your captions should include key elements of the image such as content, style, etc. As well as background content and any other important captions. if the image has an obvious style or filter, it needs to be captioned as well.your captions should be accurate and not repetitive. The caption will be used for image reconstruction, so the more similar to the original image, the better the quality of the captions. Special captions will be rewarded with $10 per image. Only the content of the best one caption will be output in the end.', image), gen_config=gen_config)
14
  return response.text
15
 
16
  css = """
 
10
  pipe = pipeline('Lin-Chen/open-llava-next-llama3-8b')
11
  gen_config = GenerationConfig(repetition_penalty=1.10)
12
  image = Image.fromarray(np.uint8(image)).convert('RGB')
13
+ response = pipe(('As an AI image annotation expert, please provide accurate annotations for the image to enhance the T5 model understanding of the content. Accurately describe images and images in the form of natural language. Your description should include key elements such as the actions, clothing, hairstyle, facial expressions, environment, dressing style, etc. of the characters in the image, as well as background content and any other important information. If the image has a distinct special style or filter, it needs to be described, otherwise it is not necessary. Your description should be accurate and accurate, only describing the actual content of the image, without describing abstract feelings such as atmosphere or quality, and should not exceed three sentences. These descriptions will be used for image reconstruction, so the closer the similarity to the original image, the better the label quality. Special tags will receive a reward of $10 per image.', image), gen_config=gen_config)
14
  return response.text
15
 
16
  css = """