Spaces:
Sleeping
Sleeping
demo file with dependency
Browse files- app.py +78 -0
- requirements.txt +7 -0
app.py
ADDED
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
import torch
|
4 |
+
from torchvision.transforms import ToTensor
|
5 |
+
from PIL import Image
|
6 |
+
|
7 |
+
# loading EfficientSAM model
|
8 |
+
model_path = "efficientsam_s_cpu.jit"
|
9 |
+
with open(model_path, "rb") as f:
|
10 |
+
model = torch.jit.load(f)
|
11 |
+
|
12 |
+
# getting mask using points
|
13 |
+
def get_sam_mask_using_points(img_tensor, pts_sampled, model):
|
14 |
+
pts_sampled = torch.reshape(torch.tensor(pts_sampled), [1, 1, -1, 2])
|
15 |
+
max_num_pts = pts_sampled.shape[2]
|
16 |
+
pts_labels = torch.ones(1, 1, max_num_pts)
|
17 |
+
|
18 |
+
predicted_logits, predicted_iou = model(
|
19 |
+
img_tensor[None, ...],
|
20 |
+
pts_sampled,
|
21 |
+
pts_labels,
|
22 |
+
)
|
23 |
+
predicted_logits = predicted_logits.cpu()
|
24 |
+
all_masks = torch.ge(torch.sigmoid(predicted_logits[0, 0, :, :, :]), 0.5).numpy()
|
25 |
+
predicted_iou = predicted_iou[0, 0, ...].cpu().detach().numpy()
|
26 |
+
|
27 |
+
max_predicted_iou = -1
|
28 |
+
selected_mask_using_predicted_iou = None
|
29 |
+
for m in range(all_masks.shape[0]):
|
30 |
+
curr_predicted_iou = predicted_iou[m]
|
31 |
+
if (
|
32 |
+
curr_predicted_iou > max_predicted_iou
|
33 |
+
or selected_mask_using_predicted_iou is None
|
34 |
+
):
|
35 |
+
max_predicted_iou = curr_predicted_iou
|
36 |
+
selected_mask_using_predicted_iou = all_masks[m]
|
37 |
+
return selected_mask_using_predicted_iou
|
38 |
+
|
39 |
+
# examples
|
40 |
+
examples = [["examples/image1.jpg"], ["examples/image2.jpg"], ["examples/image3.jpg"], ["examples/image4.jpg"],
|
41 |
+
["examples/image5.jpg"], ["examples/image6.jpg"], ["examples/image7.jpg"], ["examples/image8.jpg"],
|
42 |
+
["examples/image9.jpg"], ["examples/image10.jpg"], ["examples/image11.jpg"], ["examples/image12.jpg"]
|
43 |
+
["examples/image13.jpg"], ["examples/image14.jpg"]]
|
44 |
+
|
45 |
+
|
46 |
+
with gr.Blocks() as demo:
|
47 |
+
with gr.Row():
|
48 |
+
input_img = gr.Image(label="Input",height=512)
|
49 |
+
output_img = gr.Image(label="Selected Segment",height=512)
|
50 |
+
|
51 |
+
with gr.Row():
|
52 |
+
gr.Markdown("Try some of the examples below ⬇️")
|
53 |
+
gr.Examples(examples=examples,
|
54 |
+
inputs=[input_img])
|
55 |
+
|
56 |
+
def get_select_coords(img, evt: gr.SelectData):
|
57 |
+
img_tensor = ToTensor()(img)
|
58 |
+
_, H, W = img_tensor.shape
|
59 |
+
|
60 |
+
visited_pixels = set()
|
61 |
+
pixels_in_queue = set()
|
62 |
+
pixels_in_segment = set()
|
63 |
+
|
64 |
+
mask = get_sam_mask_using_points(img_tensor, [[evt.index[0], evt.index[1]]], model)
|
65 |
+
|
66 |
+
out = img.copy()
|
67 |
+
|
68 |
+
out = out.astype(np.uint8)
|
69 |
+
out *= mask[:,:,None]
|
70 |
+
for pixel in pixels_in_segment:
|
71 |
+
out[pixel[0], pixel[1]] = img[pixel[0], pixel[1]]
|
72 |
+
print(out)
|
73 |
+
return out
|
74 |
+
|
75 |
+
input_img.select(get_select_coords, [input_img], output_img)
|
76 |
+
|
77 |
+
if __name__ == "__main__":
|
78 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch==2.0.1
|
2 |
+
torchvision==0.15.2
|
3 |
+
gradio
|
4 |
+
transformers==4.32.0
|
5 |
+
opencv-python
|
6 |
+
pandas==2.0.3
|
7 |
+
matplotlib==3.7.2
|