Update app.py
Browse files
app.py
CHANGED
@@ -1,19 +1,20 @@
|
|
1 |
-
from transformers import CLIPSegProcessor, CLIPSegForImageSegmentation
|
2 |
import gradio as gr
|
3 |
from PIL import Image
|
4 |
import torch
|
5 |
import matplotlib.pyplot as plt
|
6 |
-
import torch
|
7 |
import numpy as np
|
|
|
|
|
8 |
|
9 |
processor = CLIPSegProcessor.from_pretrained("CIDAS/clipseg-rd64-refined")
|
10 |
-
model = CLIPSegForImageSegmentation.from_pretrained("CIDAS/clipseg-rd64-refined")
|
11 |
-
|
12 |
|
|
|
13 |
def process_image(image, prompt):
|
14 |
inputs = processor(
|
15 |
text=prompt, images=image, padding="max_length", return_tensors="pt"
|
16 |
)
|
|
|
17 |
|
18 |
# predict
|
19 |
with torch.no_grad():
|
@@ -33,22 +34,22 @@ def process_image(image, prompt):
|
|
33 |
mask = (mask - mask_min) / (mask_max - mask_min)
|
34 |
return mask
|
35 |
|
36 |
-
|
37 |
-
def get_masks(prompts, img,
|
38 |
prompts = prompts.split(",")
|
39 |
masks = []
|
40 |
for prompt in prompts:
|
41 |
mask = process_image(img, prompt)
|
42 |
-
mask = mask >
|
43 |
masks.append(mask)
|
44 |
return masks
|
45 |
|
46 |
-
|
47 |
-
def extract_image(pos_prompts, neg_prompts, img,
|
48 |
positive_masks = get_masks(pos_prompts, img, 0.5)
|
49 |
negative_masks = get_masks(neg_prompts, img, 0.5)
|
50 |
|
51 |
-
# combine masks into one
|
52 |
pos_mask = np.any(np.stack(positive_masks), axis=0)
|
53 |
neg_mask = np.any(np.stack(negative_masks), axis=0)
|
54 |
final_mask = pos_mask & ~neg_mask
|
@@ -59,12 +60,10 @@ def extract_image(pos_prompts, neg_prompts, img, threhsold):
|
|
59 |
output_image.paste(img, mask=final_mask)
|
60 |
return output_image, final_mask
|
61 |
|
62 |
-
|
63 |
title = "Interactive demo: zero-shot image segmentation with CLIPSeg"
|
64 |
description = "Demo for using CLIPSeg, a CLIP-based model for zero- and one-shot image segmentation. To use it, simply upload an image and add a text to mask (identify in the image), or use one of the examples below and click 'submit'. Results will show up in a few seconds."
|
65 |
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2112.10003'>CLIPSeg: Image Segmentation Using Text and Image Prompts</a> | <a href='https://huggingface.co/docs/transformers/main/en/model_doc/clipseg'>HuggingFace docs</a></p>"
|
66 |
|
67 |
-
|
68 |
with gr.Blocks() as demo:
|
69 |
gr.Markdown("# CLIPSeg: Image Segmentation Using Text and Image Prompts")
|
70 |
gr.Markdown(article)
|
@@ -100,5 +99,4 @@ with gr.Blocks() as demo:
|
|
100 |
outputs=[output_image, output_mask],
|
101 |
)
|
102 |
|
103 |
-
|
104 |
demo.launch()
|
|
|
|
|
1 |
import gradio as gr
|
2 |
from PIL import Image
|
3 |
import torch
|
4 |
import matplotlib.pyplot as plt
|
|
|
5 |
import numpy as np
|
6 |
+
import spaces
|
7 |
+
from transformers import CLIPSegProcessor, CLIPSegForImageSegmentation
|
8 |
|
9 |
processor = CLIPSegProcessor.from_pretrained("CIDAS/clipseg-rd64-refined")
|
10 |
+
model = CLIPSegForImageSegmentation.from_pretrained("CIDAS/clipseg-rd64-refined").cuda()
|
|
|
11 |
|
12 |
+
@spaces.GPU
|
13 |
def process_image(image, prompt):
|
14 |
inputs = processor(
|
15 |
text=prompt, images=image, padding="max_length", return_tensors="pt"
|
16 |
)
|
17 |
+
inputs = {k: v.cuda() for k, v in inputs.items()}
|
18 |
|
19 |
# predict
|
20 |
with torch.no_grad():
|
|
|
34 |
mask = (mask - mask_min) / (mask_max - mask_min)
|
35 |
return mask
|
36 |
|
37 |
+
@spaces.GPU
|
38 |
+
def get_masks(prompts, img, threshold):
|
39 |
prompts = prompts.split(",")
|
40 |
masks = []
|
41 |
for prompt in prompts:
|
42 |
mask = process_image(img, prompt)
|
43 |
+
mask = mask > threshold
|
44 |
masks.append(mask)
|
45 |
return masks
|
46 |
|
47 |
+
@spaces.GPU
|
48 |
+
def extract_image(pos_prompts, neg_prompts, img, threshold):
|
49 |
positive_masks = get_masks(pos_prompts, img, 0.5)
|
50 |
negative_masks = get_masks(neg_prompts, img, 0.5)
|
51 |
|
52 |
+
# combine masks into one mask, logic OR
|
53 |
pos_mask = np.any(np.stack(positive_masks), axis=0)
|
54 |
neg_mask = np.any(np.stack(negative_masks), axis=0)
|
55 |
final_mask = pos_mask & ~neg_mask
|
|
|
60 |
output_image.paste(img, mask=final_mask)
|
61 |
return output_image, final_mask
|
62 |
|
|
|
63 |
title = "Interactive demo: zero-shot image segmentation with CLIPSeg"
|
64 |
description = "Demo for using CLIPSeg, a CLIP-based model for zero- and one-shot image segmentation. To use it, simply upload an image and add a text to mask (identify in the image), or use one of the examples below and click 'submit'. Results will show up in a few seconds."
|
65 |
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2112.10003'>CLIPSeg: Image Segmentation Using Text and Image Prompts</a> | <a href='https://huggingface.co/docs/transformers/main/en/model_doc/clipseg'>HuggingFace docs</a></p>"
|
66 |
|
|
|
67 |
with gr.Blocks() as demo:
|
68 |
gr.Markdown("# CLIPSeg: Image Segmentation Using Text and Image Prompts")
|
69 |
gr.Markdown(article)
|
|
|
99 |
outputs=[output_image, output_mask],
|
100 |
)
|
101 |
|
|
|
102 |
demo.launch()
|