File size: 2,957 Bytes
5644c15
 
 
f3989da
5644c15
 
 
 
 
 
 
 
 
 
 
 
 
 
 
65bd2e3
 
 
 
 
 
 
 
dad7590
65bd2e3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dad7590
65bd2e3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
from transformers import Blip2Processor, Blip2ForConditionalGeneration
import torch
from PIL import Image
from typing import Optional, List, Tuple 

class ImageCaptioner:
    def __init__(self, model_name: str = "Salesforce/blip2-opt-2.7b"):
        """
        Initialize BLIP-2 model for image captioning
        
        Args:
            model_name: HuggingFace model identifier for BLIP-2
        """
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.processor = Blip2Processor.from_pretrained(model_name)
        self.model = Blip2ForConditionalGeneration.from_pretrained(
            model_name, 
            torch_dtype=torch.float16 if self.device == "cuda" else torch.float32
        ).to(self.device)


    def generate_caption(
        self, 
        image: Image.Image, 
        prompt: Optional[str] = None,
        max_length: int = 50,
        num_beams: int = 5
        
    ) -> str:
        """
        Generate caption for an image
        
        Args:
            image: PIL Image to caption
            prompt: Optional prompt to guide caption generation
            max_length: Maximum length of generated caption
            num_beams: Number of beams for beam search
            
        Returns:
            Generated caption string
        """
        # Prepare image for the model
        inputs = self.processor(
            images=image,
            text=prompt if prompt else "a photo of",
            return_tensors="pt"
        ).to(self.device)
        
        # Generate caption
        with torch.no_grad():
            generated_ids = self.model.generate(
                **inputs,
                max_length=max_length,
                num_beams=num_beams,
                min_length=10,
                do_sample=True, 
                top_p=0.9,
                repetition_penalty=1.5
            )
        
        # Decode the generated caption
        generated_text = self.processor.batch_decode(
            generated_ids, 
            skip_special_tokens=True
        )[0].strip()
        
        return generated_text

    def batch_generate_captions(
        self,
        images: List[Image.Image],
        prompt: Optional[str] = None,
        batch_size: int = 4
    ) -> List[str]:
        """
        Generate captions for multiple images
        
        Args:
            images: List of PIL Images
            prompt: Optional prompt for all images
            batch_size: Number of images to process at once
            
        Returns:
            List of generated captions
        """
        captions = []
        
        # Process images in batches
        for i in range(0, len(images), batch_size):
            batch = images[i:i + batch_size]
            batch_captions = [
                self.generate_caption(img, prompt)
                for img in batch
            ]
            captions.extend(batch_captions)
        
        return captions