Spaces:
Configuration error
Configuration error
Update README.md
Browse files
README.md
CHANGED
@@ -13,6 +13,72 @@ lat_std = 0.0006361722351128644 \
|
|
13 |
lon_mean = -75.19150880602636 \
|
14 |
lon_std = 0.000611411894337979
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
The model can be loaded using:
|
17 |
```
|
18 |
from huggingface_hub import hf_hub_download
|
|
|
13 |
lon_mean = -75.19150880602636 \
|
14 |
lon_std = 0.000611411894337979
|
15 |
|
16 |
+
The model implementation is found here:
|
17 |
+
```
|
18 |
+
import torch
|
19 |
+
import torch.nn as nn
|
20 |
+
import torchvision.models as models
|
21 |
+
import torchvision.transforms as transforms
|
22 |
+
from torch.utils.data import DataLoader, Dataset
|
23 |
+
from transformers import AutoImageProcessor, AutoModelForImageClassification
|
24 |
+
from huggingface_hub import PyTorchModelHubMixin
|
25 |
+
from PIL import Image
|
26 |
+
import os
|
27 |
+
import numpy as np
|
28 |
+
from transformers import AutoModel
|
29 |
+
|
30 |
+
class MultiModalModel(nn.Module):
|
31 |
+
def __init__(self, image_model_name="google/vit-base-patch16-224", num_gps_features=2, output_dim=2):
|
32 |
+
super(MultiModalModel, self).__init__()
|
33 |
+
# Load Vision Transformer for feature extraction
|
34 |
+
self.image_model = AutoModel.from_pretrained(image_model_name, output_hidden_states=True)
|
35 |
+
|
36 |
+
# Reduce image features to match GPS features
|
37 |
+
self.image_fc = nn.Sequential(
|
38 |
+
nn.Linear(self.image_model.config.hidden_size, 256),
|
39 |
+
nn.ReLU(),
|
40 |
+
)
|
41 |
+
|
42 |
+
# Process GPS features
|
43 |
+
self.gps_fc = nn.Sequential(
|
44 |
+
nn.Linear(num_gps_features, 128),
|
45 |
+
nn.ReLU(),
|
46 |
+
nn.Dropout(0.3),
|
47 |
+
nn.Linear(128, 256),
|
48 |
+
)
|
49 |
+
|
50 |
+
# Combine image and GPS features for regression
|
51 |
+
self.regressor = nn.Sequential(
|
52 |
+
nn.Linear(256 + 256, 512), # 256 from image + 256 from GPS
|
53 |
+
nn.ReLU(),
|
54 |
+
nn.Dropout(0.4),
|
55 |
+
nn.Linear(512, output_dim),
|
56 |
+
)
|
57 |
+
|
58 |
+
def forward(self, image, gps):
|
59 |
+
# Extract image features from the last hidden state
|
60 |
+
image_outputs = self.image_model(image)
|
61 |
+
image_features = image_outputs.last_hidden_state[:, 0, :] # CLS token features
|
62 |
+
image_features = self.image_fc(image_features)
|
63 |
+
|
64 |
+
# Process GPS features
|
65 |
+
gps_features = self.gps_fc(gps)
|
66 |
+
|
67 |
+
# Concatenate image and GPS features
|
68 |
+
combined_features = torch.cat([image_features, gps_features], dim=1)
|
69 |
+
|
70 |
+
# Final regression
|
71 |
+
return self.regressor(combined_features)
|
72 |
+
|
73 |
+
def save_model(self, save_path):
|
74 |
+
"""Save model locally using the Hugging Face format."""
|
75 |
+
self.save_pretrained(save_path)
|
76 |
+
|
77 |
+
def push_model(self, repo_name):
|
78 |
+
"""Push the model to the Hugging Face Hub."""
|
79 |
+
self.push_to_hub(repo_name)
|
80 |
+
```
|
81 |
+
|
82 |
The model can be loaded using:
|
83 |
```
|
84 |
from huggingface_hub import hf_hub_download
|