Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,45 +1,45 @@
|
|
1 |
-
from transformers import VisionEncoderDecoderModel, ViTImageProcessor, AutoTokenizer
|
2 |
-
import torch
|
3 |
-
from PIL import Image
|
4 |
-
import gradio as gr
|
5 |
-
|
6 |
-
model_name = "aryan083/vit-gpt2-image-captioning"
|
7 |
-
model = VisionEncoderDecoderModel.from_pretrained(model_name)
|
8 |
-
feature_extractor =
|
9 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
10 |
-
|
11 |
-
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
12 |
-
model.to(device)
|
13 |
-
|
14 |
-
def predict_caption(image):
|
15 |
-
if image is None:
|
16 |
-
return None
|
17 |
-
|
18 |
-
images = []
|
19 |
-
images.append(image)
|
20 |
-
|
21 |
-
pixel_values = feature_extractor(images=images, return_tensors="pt").pixel_values
|
22 |
-
pixel_values = pixel_values.to(device)
|
23 |
-
|
24 |
-
output_ids = model.generate(
|
25 |
-
pixel_values,
|
26 |
-
do_sample=True,
|
27 |
-
max_length=16,
|
28 |
-
num_beams=4,
|
29 |
-
temperature=0.7
|
30 |
-
)
|
31 |
-
|
32 |
-
preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
|
33 |
-
return preds[0].strip()
|
34 |
-
|
35 |
-
# Create Gradio interface
|
36 |
-
iface = gr.Interface(
|
37 |
-
fn=predict_caption,
|
38 |
-
inputs=gr.Image(type="pil"),
|
39 |
-
outputs=gr.Textbox(label="Generated Caption"),
|
40 |
-
title="Image Captioning",
|
41 |
-
description="Upload an image and get its description generated using ViT-GPT2",
|
42 |
-
# examples=[["assets/example1.jpg"]] # Add example images if you have any
|
43 |
-
)
|
44 |
-
|
45 |
iface.launch()
|
|
|
1 |
+
from transformers import VisionEncoderDecoderModel, ViTImageProcessor, AutoTokenizer
|
2 |
+
import torch
|
3 |
+
from PIL import Image
|
4 |
+
import gradio as gr
|
5 |
+
|
6 |
+
model_name = "aryan083/vit-gpt2-image-captioning"
|
7 |
+
model = VisionEncoderDecoderModel.from_pretrained(model_name)
|
8 |
+
feature_extractor = ViTImageProcessor.from_pretrained(model_name) # Changed from ViTFeatureExtractor to ViTImageProcessor
|
9 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
10 |
+
|
11 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
12 |
+
model.to(device)
|
13 |
+
|
14 |
+
def predict_caption(image):
|
15 |
+
if image is None:
|
16 |
+
return None
|
17 |
+
|
18 |
+
images = []
|
19 |
+
images.append(image)
|
20 |
+
|
21 |
+
pixel_values = feature_extractor(images=images, return_tensors="pt").pixel_values
|
22 |
+
pixel_values = pixel_values.to(device)
|
23 |
+
|
24 |
+
output_ids = model.generate(
|
25 |
+
pixel_values,
|
26 |
+
do_sample=True,
|
27 |
+
max_length=16,
|
28 |
+
num_beams=4,
|
29 |
+
temperature=0.7
|
30 |
+
)
|
31 |
+
|
32 |
+
preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
|
33 |
+
return preds[0].strip()
|
34 |
+
|
35 |
+
# Create Gradio interface
|
36 |
+
iface = gr.Interface(
|
37 |
+
fn=predict_caption,
|
38 |
+
inputs=gr.Image(type="pil"),
|
39 |
+
outputs=gr.Textbox(label="Generated Caption"),
|
40 |
+
title="Image Captioning",
|
41 |
+
description="Upload an image and get its description generated using ViT-GPT2",
|
42 |
+
# examples=[["assets/example1.jpg"]] # Add example images if you have any
|
43 |
+
)
|
44 |
+
|
45 |
iface.launch()
|