patrickramos
commited on
Commit
·
617c3e2
1
Parent(s):
fc3d6cc
Update README.md
Browse files
README.md
CHANGED
@@ -5,6 +5,7 @@ tags:
|
|
5 |
- sklearn
|
6 |
- skops
|
7 |
- tabular-classification
|
|
|
8 |
---
|
9 |
|
10 |
# Model description
|
@@ -71,9 +72,40 @@ Use the code below to get started with the model.
|
|
71 |
<summary> Click to expand </summary>
|
72 |
|
73 |
```python
|
74 |
-
import
|
75 |
-
|
76 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
```
|
78 |
|
79 |
</details>
|
|
|
5 |
- sklearn
|
6 |
- skops
|
7 |
- tabular-classification
|
8 |
+
- visual emb-gam
|
9 |
---
|
10 |
|
11 |
# Model description
|
|
|
72 |
<summary> Click to expand </summary>
|
73 |
|
74 |
```python
|
75 |
+
from PIL import Image
|
76 |
+
from skops import hub_utils
|
77 |
+
import torch
|
78 |
+
from transformers import AutoFeatureExtractor, AutoModel
|
79 |
+
import pickle
|
80 |
+
import os
|
81 |
+
|
82 |
+
# load embedding model
|
83 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
84 |
+
feature_extractor = AutoFeatureExtractor.from_pretrained('microsoft/resnet-50')
|
85 |
+
model = AutoModel.from_pretrained('microsoft/resnet-50').eval().to(device)
|
86 |
+
|
87 |
+
# load logistic regression
|
88 |
+
os.mkdir('emb-gam-resnet')
|
89 |
+
hub_utils.download(repo_id='Ramos-Ramos/emb-gam-resnet', dst='emb-gam-resnet')
|
90 |
+
|
91 |
+
with open('emb-gam-resnet/model.pkl', 'rb') as file:
|
92 |
+
logistic_regression = pickle.load(file)
|
93 |
+
|
94 |
+
# load image
|
95 |
+
img = Image.open('examples/english_springer.png')
|
96 |
+
|
97 |
+
# preprocess image
|
98 |
+
inputs = {k: v.to(device) for k, v in feature_extractor(img, return_tensors='pt').items()}
|
99 |
+
|
100 |
+
# extract patch embeddings
|
101 |
+
with torch.no_grad():
|
102 |
+
patch_embeddings = rearrange(model(**inputs).last_hidden_state, 'b d h w -> b (h w) d').cpu()
|
103 |
+
|
104 |
+
# classify
|
105 |
+
pred = logistic_regression.predict(patch_embeddings.sum(dim=0, keepdim=True))
|
106 |
+
|
107 |
+
# get patch contributions
|
108 |
+
patch_contributions = logistic_regression.coef_ @ patch_embeddings.T.numpy()
|
109 |
```
|
110 |
|
111 |
</details>
|