File size: 11,334 Bytes
eb1aec4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295

"""
SatCLIP Multi-Spectral Model

This model supports both RGB input and full multi-spectral Sentinel-2 input (12/13 bands).
"""
import sys
import os
import torch
import numpy as np
import pandas as pd
import pyarrow.parquet as pq
import torch.nn.functional as F
import torchvision.transforms as T
from PIL import Image
from huggingface_hub import hf_hub_download

import warnings

with warnings.catch_warnings():
    warnings.filterwarnings("ignore", category=FutureWarning)
    try:
        from models.SatCLIP.satclip.load import get_satclip
        print("Successfully imported models.SatCLIP.satclip.load.get_satclip.")
    except ImportError:
        get_satclip = None
        print("Warning: SatCLIP not available. Please check installation.")


class SatCLIPMSModel:
    """
    SatCLIP model wrapper supporting multi-spectral Sentinel-2 input.
    
    Supports:
    - RGB PIL Image input (auto-converted to 13 channels)
    - 12-band Sentinel-2 numpy array (auto-padded to 13 channels)
    - 13-band full Sentinel-2 tensor
    """
    
    def __init__(self, 
                 ckpt_path='./checkpoints/SatCLIP/satclip-vit16-l40.ckpt',
                 embedding_path=None,
                 device=None):
        
        self.device = device if device else ("cuda" if torch.cuda.is_available() else "cpu")
        if ckpt_path and 'hf' in ckpt_path:
            ckpt_path = hf_hub_download("microsoft/SatCLIP-ViT16-L40", "satclip-vit16-l40.ckpt")
        self.ckpt_path = ckpt_path
        self.embedding_path = embedding_path
        
        self.model = None
        self.df_embed = None
        self.image_embeddings = None
        
        # SatCLIP input size
        self.input_size = 224
        
        # Sentinel-2 bands mapping
        # MajorTOM 12 bands: [B01, B02, B03, B04, B05, B06, B07, B08, B8A, B09, B11, B12]
        # SatCLIP 13 bands:  [B01, B02, B03, B04, B05, B06, B07, B08, B8A, B09, B10, B11, B12]
        # B10 is missing in MajorTOM, need to insert zeros at index 10
        self.majortom_bands = ['B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09', 'B11', 'B12']
        self.satclip_bands = ['B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09', 'B10', 'B11', 'B12']
        
        self.load_model()
        if self.embedding_path:
            self.load_embeddings()

    def load_model(self):
        if get_satclip is None:
            print("Error: SatCLIP functionality is not available.")
            return

        print(f"Loading SatCLIP-MS model from {self.ckpt_path}...")
        try:
            if not os.path.exists(self.ckpt_path):
                print(f"Warning: Checkpoint not found at {self.ckpt_path}")
                return

            # Load model using get_satclip
            # return_all=True to get both visual and location encoders
            self.model = get_satclip(self.ckpt_path, self.device, return_all=True)
            self.model.eval()
            print(f"SatCLIP-MS model loaded on {self.device}")
        except Exception as e:
            print(f"Error loading SatCLIP model: {e}")

    def load_embeddings(self):
        print(f"Loading SatCLIP embeddings from {self.embedding_path}...")
        try:
            if not os.path.exists(self.embedding_path):
                print(f"Warning: Embedding file not found at {self.embedding_path}")
                return

            self.df_embed = pq.read_table(self.embedding_path).to_pandas()
            
            # Pre-compute image embeddings tensor
            image_embeddings_np = np.stack(self.df_embed['embedding'].values)
            self.image_embeddings = torch.from_numpy(image_embeddings_np).to(self.device).float()
            self.image_embeddings = F.normalize(self.image_embeddings, dim=-1)
            print(f"SatCLIP Data loaded: {len(self.df_embed)} records")
        except Exception as e:
            print(f"Error loading SatCLIP embeddings: {e}")

    def encode_location(self, lat, lon):
        """
        Encode a (latitude, longitude) pair into a vector.
        """
        if self.model is None:
            return None
        
        # SatCLIP expects input shape (N, 2) -> (lon, lat)
        coords = torch.tensor([[lon, lat]], dtype=torch.double).to(self.device)
        
        with torch.no_grad():
            loc_features = self.model.encode_location(coords).float()
            loc_features = loc_features / loc_features.norm(dim=1, keepdim=True)
            
        return loc_features

    def _preprocess_12band_array(self, img_array: np.ndarray) -> torch.Tensor:
        """
        Preprocess a 12-band Sentinel-2 array to 13-band tensor for SatCLIP.
        
        Args:
            img_array: numpy array of shape (H, W, 12) with uint16 values (0-10000+)
            
        Returns:
            torch.Tensor of shape (13, 224, 224) normalized
        """
        # 1. Normalize (SatCLIP standard: / 10000.0)
        image = img_array.astype(np.float32) / 10000.0
        
        # 2. Channel First: (H, W, C) -> (C, H, W) -> (12, H, W)
        image = image.transpose(2, 0, 1)
        
        # 3. Insert B10 (zeros) at index 10 -> (13, H, W)
        # MajorTOM: [B01..B09(idx0-9), B11(idx10), B12(idx11)]
        # SatCLIP:  [B01..B09(idx0-9), B10(idx10), B11(idx11), B12(idx12)]
        B10 = np.zeros((1, image.shape[1], image.shape[2]), dtype=image.dtype)
        image_13 = np.concatenate([image[:10], B10, image[10:]], axis=0)
        
        # 4. Convert to Tensor
        image_tensor = torch.from_numpy(image_13)
        
        # 5. Resize to 224x224
        transforms = T.Resize((self.input_size, self.input_size), 
                             interpolation=T.InterpolationMode.BICUBIC,
                             antialias=True)
        image_tensor = transforms(image_tensor)
        
        return image_tensor

    def _preprocess_rgb_image(self, image: Image.Image) -> torch.Tensor:
        """
        Preprocess RGB PIL Image to 13-band tensor for SatCLIP.
        Maps RGB to B04, B03, B02 and zeros for other bands.
        
        Args:
            image: PIL RGB Image
            
        Returns:
            torch.Tensor of shape (13, 224, 224)
        """
        image = image.convert("RGB")
        image = image.resize((self.input_size, self.input_size))
        img_np = np.array(image).astype(np.float32) / 255.0
        
        # Construct 13 channels
        # S2 bands: B01, B02(B), B03(G), B04(R), B05...
        # Indices:  0=B01, 1=B02, 2=B03, 3=B04 ...
        input_tensor = np.zeros((13, self.input_size, self.input_size), dtype=np.float32)
        input_tensor[1] = img_np[:, :, 2]  # Blue -> B02
        input_tensor[2] = img_np[:, :, 1]  # Green -> B03
        input_tensor[3] = img_np[:, :, 0]  # Red -> B04
        
        return torch.from_numpy(input_tensor)

    def encode_image(self, image, is_multispectral=False):
        """
        Encode an image into a feature vector.
        
        Args:
            image: Can be one of:
                - PIL.Image (RGB) - will be converted to 13-band
                - np.ndarray of shape (H, W, 12) - 12-band Sentinel-2 data
                - torch.Tensor of shape (13, H, W) or (B, 13, H, W) - ready for model
            is_multispectral: Hint to indicate if numpy input is multi-spectral
            
        Returns:
            torch.Tensor: Normalized embedding vector
        """
        if self.model is None:
            return None
        
        try:
            # Handle different input types
            if isinstance(image, Image.Image):
                # RGB PIL Image
                input_tensor = self._preprocess_rgb_image(image).unsqueeze(0)
                
            elif isinstance(image, np.ndarray):
                # Numpy array - assumed to be 12-band Sentinel-2 (H, W, 12)
                if image.ndim == 3 and image.shape[-1] == 12:
                    input_tensor = self._preprocess_12band_array(image).unsqueeze(0)
                elif image.ndim == 3 and image.shape[-1] == 3:
                    # RGB numpy array
                    pil_img = Image.fromarray(image.astype(np.uint8))
                    input_tensor = self._preprocess_rgb_image(pil_img).unsqueeze(0)
                else:
                    print(f"Unsupported numpy array shape: {image.shape}")
                    return None
                    
            elif isinstance(image, torch.Tensor):
                # Already a tensor
                if image.dim() == 3:
                    input_tensor = image.unsqueeze(0)
                else:
                    input_tensor = image
                    
                # Resize if needed
                if input_tensor.shape[-1] != self.input_size or input_tensor.shape[-2] != self.input_size:
                    transforms = T.Resize((self.input_size, self.input_size),
                                         interpolation=T.InterpolationMode.BICUBIC,
                                         antialias=True)
                    input_tensor = transforms(input_tensor)
            else:
                print(f"Unsupported image type: {type(image)}")
                return None
            
            # Move to device and encode
            input_tensor = input_tensor.to(self.device)
            
            with torch.no_grad():
                img_feature = self.model.encode_image(input_tensor)
                img_feature = img_feature / img_feature.norm(dim=1, keepdim=True)
                
            return img_feature
            
        except Exception as e:
            print(f"Error encoding image in SatCLIP-MS: {e}")
            import traceback
            traceback.print_exc()
            return None

    def encode_batch(self, batch_tensors: list) -> np.ndarray:
        """
        Encode a batch of pre-processed tensors.
        
        Args:
            batch_tensors: List of torch.Tensor, each of shape (13, H, W)
            
        Returns:
            np.ndarray of shape (N, embedding_dim)
        """
        if self.model is None:
            return None
            
        try:
            t_stack = torch.stack(batch_tensors).to(self.device)
            
            with torch.no_grad():
                feats = self.model.encode_image(t_stack)
                feats = feats / feats.norm(dim=1, keepdim=True)
                
            return feats.cpu().numpy()
            
        except Exception as e:
            print(f"Error encoding batch: {e}")
            return None

    def search(self, query_features, top_k=5, top_percent=None, threshold=0.0):
        if self.image_embeddings is None:
            return None, None, None

        query_features = query_features.float()
        
        # Similarity calculation (Cosine similarity)
        probs = (self.image_embeddings @ query_features.T).detach().cpu().numpy().flatten()
        
        if top_percent is not None:
            k = int(len(probs) * top_percent)
            if k < 1:
                k = 1
            threshold = np.partition(probs, -k)[-k]

        # Filter by threshold
        mask = probs >= threshold
        filtered_indices = np.where(mask)[0]
        
        # Get top k
        top_indices = np.argsort(probs)[-top_k:][::-1]
        
        return probs, filtered_indices, top_indices