File size: 6,806 Bytes
7a3ba2e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
"""
OWLv2 Custom Handler for HuggingFace Inference Endpoints

Supports:
- Image-conditioned detection (find objects similar to a reference image)
- Text-conditioned detection (find objects matching text descriptions)
"""

from typing import Dict, Any, List, Union
import torch
from transformers import Owlv2Processor, Owlv2ForObjectDetection
from PIL import Image
import base64
import io


class EndpointHandler:
    def __init__(self, path=""):
        """Load model on endpoint startup."""
        model_id = "google/owlv2-large-patch14-ensemble"
        
        self.processor = Owlv2Processor.from_pretrained(model_id)
        self.model = Owlv2ForObjectDetection.from_pretrained(model_id)
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.model = self.model.to(self.device)
        self.model.eval()
        
        print(f"OWLv2 loaded on {self.device}")

    def _decode_image(self, image_data: str) -> Image.Image:
        """Decode base64 image string to PIL Image."""
        # Handle data URL format (e.g., "data:image/jpeg;base64,...")
        if "," in image_data:
            image_data = image_data.split(",")[1]
        
        image_bytes = base64.b64decode(image_data)
        image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
        return image

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """
        Process detection request.
        
        === Image-Conditioned Detection ===
        Find objects similar to a reference image.
        
        Request:
        {
            "inputs": {
                "target_image": "base64...",
                "query_image": "base64...",
                "threshold": 0.5,
                "nms_threshold": 0.3
            }
        }
        
        === Text-Conditioned Detection ===
        Find objects matching text descriptions.
        
        Request:
        {
            "inputs": {
                "target_image": "base64...",
                "queries": ["a button", "an icon"],
                "threshold": 0.1
            }
        }
        
        === Multiple Query Images ===
        Find multiple different objects by image.
        
        Request:
        {
            "inputs": {
                "target_image": "base64...",
                "query_images": ["base64...", "base64..."],
                "threshold": 0.5,
                "nms_threshold": 0.3
            }
        }
        
        Response:
        {
            "detections": [
                {"box": [x1, y1, x2, y2], "confidence": 0.95, "label": "query_0"}
            ]
        }
        """
        try:
            # Handle both {"inputs": {...}} and direct {...} format
            inputs = data.get("inputs", data)
            
            # Validate required field
            if "target_image" not in inputs:
                return {"error": "Missing required field: target_image"}
            
            target_image = self._decode_image(inputs["target_image"])
            threshold = float(inputs.get("threshold", 0.5))
            nms_threshold = float(inputs.get("nms_threshold", 0.3))
            
            # Route to appropriate detection method
            if "query_image" in inputs:
                # Single query image
                query_image = self._decode_image(inputs["query_image"])
                return self._detect_with_image(
                    target_image, [query_image], threshold, nms_threshold
                )
            
            elif "query_images" in inputs:
                # Multiple query images
                query_images = [
                    self._decode_image(img) for img in inputs["query_images"]
                ]
                return self._detect_with_image(
                    target_image, query_images, threshold, nms_threshold
                )
            
            elif "queries" in inputs:
                # Text queries
                return self._detect_with_text(
                    target_image, inputs["queries"], threshold
                )
            
            else:
                return {
                    "error": "Provide 'query_image', 'query_images', or 'queries'"
                }
                
        except Exception as e:
            return {"error": str(e)}

    def _detect_with_image(
        self,
        target: Image.Image,
        query_images: List[Image.Image],
        threshold: float,
        nms_threshold: float
    ) -> Dict[str, Any]:
        """Image-conditioned detection."""
        
        inputs = self.processor(
            images=target,
            query_images=query_images,
            return_tensors="pt"
        )
        inputs = {k: v.to(self.device) for k, v in inputs.items()}
        
        with torch.no_grad():
            outputs = self.model.image_guided_detection(**inputs)
        
        target_sizes = torch.tensor([target.size[::-1]])  # (height, width)
        results = self.processor.post_process_image_guided_detection(
            outputs=outputs,
            threshold=threshold,
            nms_threshold=nms_threshold,
            target_sizes=target_sizes
        )[0]
        
        detections = []
        for i, (box, score) in enumerate(zip(results["boxes"], results["scores"])):
            det = {
                "box": [round(c, 2) for c in box.tolist()],
                "confidence": round(score.item(), 4)
            }
            # Add label if multiple query images
            if len(query_images) > 1 and "labels" in results:
                det["label"] = f"query_{results['labels'][i].item()}"
            detections.append(det)
        
        return {"detections": detections}

    def _detect_with_text(
        self,
        target: Image.Image,
        queries: List[str],
        threshold: float
    ) -> Dict[str, Any]:
        """Text-conditioned detection."""
        
        inputs = self.processor(
            text=[queries],
            images=target,
            return_tensors="pt"
        )
        inputs = {k: v.to(self.device) for k, v in inputs.items()}
        
        with torch.no_grad():
            outputs = self.model(**inputs)
        
        target_sizes = torch.tensor([target.size[::-1]])
        results = self.processor.post_process_object_detection(
            outputs, threshold=threshold, target_sizes=target_sizes
        )[0]
        
        detections = []
        for box, score, label_idx in zip(
            results["boxes"], results["scores"], results["labels"]
        ):
            detections.append({
                "box": [round(c, 2) for c in box.tolist()],
                "confidence": round(score.item(), 4),
                "label": queries[label_idx.item()]
            })
        
        return {"detections": detections}