Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from huggingface_hub import hf_hub_download
|
3 |
+
import torch, open_clip
|
4 |
+
from PIL import Image
|
5 |
+
from IPython.display import display
|
6 |
+
|
7 |
+
for model_name in ['RN50', 'ViT-B-32', 'ViT-L-14']:
|
8 |
+
checkpoint_path = hf_hub_download("chendelong/RemoteCLIP", f"RemoteCLIP-{model_name}.pt", cache_dir='checkpoints')
|
9 |
+
print(f'{model_name} is downloaded to {checkpoint_path}.')
|
10 |
+
|
11 |
+
model_name = 'RN50' # 'RN50' or 'ViT-B-32' or 'ViT-L-14'
|
12 |
+
model, _, preprocess = open_clip.create_model_and_transforms(model_name)
|
13 |
+
tokenizer = open_clip.get_tokenizer(model_name)
|
14 |
+
|
15 |
+
path_to_your_checkpoints = 'checkpoints/models--chendelong--RemoteCLIP/snapshots/bf1d8a3ccf2ddbf7c875705e46373bfe542bce38'
|
16 |
+
|
17 |
+
ckpt = torch.load(f"{path_to_your_checkpoints}/RemoteCLIP-{model_name}.pt", map_location="cpu")
|
18 |
+
|
19 |
+
def remote_clip(input_image,input_text):
|
20 |
+
|
21 |
+
text_queries = [input_text]
|
22 |
+
text = tokenizer(text_queries)
|
23 |
+
|
24 |
+
image = Image.open(input_image)
|
25 |
+
image = preprocess(image).unsqueeze(0)
|
26 |
+
|
27 |
+
with torch.no_grad(), torch.cuda.amp.autocast():
|
28 |
+
image_features = model.encode_image(image.cuda())
|
29 |
+
text_features = model.encode_text(text.cuda())
|
30 |
+
image_features /= image_features.norm(dim=-1, keepdim=True)
|
31 |
+
text_features /= text_features.norm(dim=-1, keepdim=True)
|
32 |
+
|
33 |
+
text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1).cpu().numpy()[0]
|
34 |
+
|
35 |
+
print(f'Predictions of {model_name}:')
|
36 |
+
for query, prob in zip(text_queries, text_probs):
|
37 |
+
print(f"{query:<40} {prob * 100:5.1f}%")
|
38 |
+
|
39 |
+
|
40 |
+
demo = gr.Interface(fn=greet, inputs=[gr.Image(type="pil"), gr.Text(type="pil")], outputs="text")
|
41 |
+
|
42 |
+
demo.launch()
|