Update app.py
Browse files
app.py
CHANGED
|
@@ -3,50 +3,50 @@ import gradio as gr
|
|
| 3 |
from transformers import CLIPProcessor, CLIPModel
|
| 4 |
import spaces
|
| 5 |
|
| 6 |
-
model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
|
| 7 |
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
|
| 8 |
|
| 9 |
-
|
| 10 |
-
@spaces.GPU(duration=120)
|
| 11 |
def calculate_score(image, text):
|
|
|
|
|
|
|
| 12 |
labels = text.split(";")
|
| 13 |
labels = [l.strip() for l in labels]
|
| 14 |
labels = list(filter(None, labels))
|
| 15 |
if len(labels) == 0:
|
| 16 |
return dict()
|
|
|
|
| 17 |
inputs = processor(text=labels, images=image, return_tensors="pt", padding=True)
|
| 18 |
-
inputs = {
|
| 19 |
-
|
| 20 |
-
}
|
| 21 |
outputs = model(**inputs)
|
| 22 |
-
logits_per_image = (
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
results_dict = {
|
| 27 |
-
label: score / 100.0 for label, score in zip(labels, logits_per_image[0])
|
| 28 |
-
}
|
| 29 |
return results_dict
|
| 30 |
|
| 31 |
-
|
| 32 |
with gr.Blocks() as demo:
|
| 33 |
gr.Markdown("# CLIP Score")
|
| 34 |
-
gr.Markdown(
|
| 35 |
-
|
| 36 |
-
)
|
| 37 |
with gr.Row():
|
| 38 |
image_input = gr.Image()
|
| 39 |
output_label = gr.Label()
|
| 40 |
-
|
| 41 |
text_input = gr.Textbox(label="Descriptions (separated by semicolons)")
|
| 42 |
-
|
| 43 |
image_input.change(
|
| 44 |
-
fn=calculate_score,
|
|
|
|
|
|
|
| 45 |
)
|
|
|
|
| 46 |
text_input.submit(
|
| 47 |
-
fn=calculate_score,
|
|
|
|
|
|
|
| 48 |
)
|
| 49 |
-
|
| 50 |
gr.Examples(
|
| 51 |
examples=[
|
| 52 |
[
|
|
@@ -59,4 +59,4 @@ with gr.Blocks() as demo:
|
|
| 59 |
outputs=output_label,
|
| 60 |
)
|
| 61 |
|
| 62 |
-
demo.launch()
|
|
|
|
| 3 |
from transformers import CLIPProcessor, CLIPModel
|
| 4 |
import spaces
|
| 5 |
|
| 6 |
+
model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
|
| 7 |
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
|
| 8 |
|
| 9 |
+
@spaces.GPU
|
|
|
|
| 10 |
def calculate_score(image, text):
|
| 11 |
+
model.to("cuda") # Move model to CUDA inside the GPU-decorated function
|
| 12 |
+
|
| 13 |
labels = text.split(";")
|
| 14 |
labels = [l.strip() for l in labels]
|
| 15 |
labels = list(filter(None, labels))
|
| 16 |
if len(labels) == 0:
|
| 17 |
return dict()
|
| 18 |
+
|
| 19 |
inputs = processor(text=labels, images=image, return_tensors="pt", padding=True)
|
| 20 |
+
inputs = {k: v.to("cuda") for k, v in inputs.items()}
|
| 21 |
+
|
|
|
|
| 22 |
outputs = model(**inputs)
|
| 23 |
+
logits_per_image = outputs.logits_per_image.detach().cpu().numpy()
|
| 24 |
+
|
| 25 |
+
results_dict = {label: score / 100.0 for label, score in zip(labels, logits_per_image[0])}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
return results_dict
|
| 27 |
|
|
|
|
| 28 |
with gr.Blocks() as demo:
|
| 29 |
gr.Markdown("# CLIP Score")
|
| 30 |
+
gr.Markdown("Calculate the [CLIP](https://openai.com/blog/clip/) score of a given image and text")
|
| 31 |
+
|
|
|
|
| 32 |
with gr.Row():
|
| 33 |
image_input = gr.Image()
|
| 34 |
output_label = gr.Label()
|
| 35 |
+
|
| 36 |
text_input = gr.Textbox(label="Descriptions (separated by semicolons)")
|
| 37 |
+
|
| 38 |
image_input.change(
|
| 39 |
+
fn=calculate_score,
|
| 40 |
+
inputs=[image_input, text_input],
|
| 41 |
+
outputs=output_label
|
| 42 |
)
|
| 43 |
+
|
| 44 |
text_input.submit(
|
| 45 |
+
fn=calculate_score,
|
| 46 |
+
inputs=[image_input, text_input],
|
| 47 |
+
outputs=output_label
|
| 48 |
)
|
| 49 |
+
|
| 50 |
gr.Examples(
|
| 51 |
examples=[
|
| 52 |
[
|
|
|
|
| 59 |
outputs=output_label,
|
| 60 |
)
|
| 61 |
|
| 62 |
+
demo.launch()
|