Spaces:
Build error
Build error
harpreetsahota
commited on
Commit
•
d13f105
1
Parent(s):
a0ae2b6
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from io import BytesIO
|
2 |
+
|
3 |
+
import cv2
|
4 |
+
import gradio as gr
|
5 |
+
import numpy as np
|
6 |
+
import requests
|
7 |
+
from PIL import Image
|
8 |
+
|
9 |
+
|
10 |
+
from super_gradients.common.object_names import Models
|
11 |
+
from super_gradients.training import models
|
12 |
+
from super_gradients.training.utils.visualization.detection import draw_bbox
|
13 |
+
from super_gradients.training.utils.visualization.pose_estimation import PoseVisualization
|
14 |
+
|
15 |
+
# Initialize your pose estimation model
|
16 |
+
yolo_nas_pose = models.get("yolo_nas_pose_l",
|
17 |
+
num_classes=17,
|
18 |
+
checkpoint_path="/content/yolo_nas_pose_l_coco_pose.pth")
|
19 |
+
|
20 |
+
def process_and_predict(url=None,
|
21 |
+
image=None,
|
22 |
+
confidence=0.5,
|
23 |
+
iou=0.5):
|
24 |
+
# If a URL is provided, use it directly for prediction
|
25 |
+
if url is not None and url.strip() != "":
|
26 |
+
response = requests.get(url)
|
27 |
+
image = Image.open(BytesIO(response.content))
|
28 |
+
image = np.array(image)
|
29 |
+
result = yolo_nas_pose.predict(image, conf=confidence,iou=iou)
|
30 |
+
# If a file is uploaded, read it, convert it to a numpy array and use it for prediction
|
31 |
+
elif image is not None:
|
32 |
+
result = yolo_nas_pose.predict(image, conf=confidence,iou=iou)
|
33 |
+
else:
|
34 |
+
return None # If no input is provided, return None
|
35 |
+
|
36 |
+
# Extract prediction data
|
37 |
+
image_prediction = result._images_prediction_lst[0]
|
38 |
+
|
39 |
+
pose_data = image_prediction.prediction
|
40 |
+
|
41 |
+
# Visualize the prediction
|
42 |
+
output_image = PoseVisualization.draw_poses(
|
43 |
+
image=image_prediction.image,
|
44 |
+
poses=pose_data.poses,
|
45 |
+
boxes=pose_data.bboxes_xyxy,
|
46 |
+
scores=pose_data.scores,
|
47 |
+
is_crowd=None,
|
48 |
+
edge_links=pose_data.edge_links,
|
49 |
+
edge_colors=pose_data.edge_colors,
|
50 |
+
keypoint_colors=pose_data.keypoint_colors,
|
51 |
+
joint_thickness=2,
|
52 |
+
box_thickness=2,
|
53 |
+
keypoint_radius=5
|
54 |
+
)
|
55 |
+
|
56 |
+
blank_image = np.zeros_like(image_prediction.image)
|
57 |
+
|
58 |
+
skeleton_image = PoseVisualization.draw_poses(
|
59 |
+
image=blank_image,
|
60 |
+
poses=pose_data.poses,
|
61 |
+
boxes=pose_data.bboxes_xyxy,
|
62 |
+
scores=pose_data.scores,
|
63 |
+
is_crowd=None,
|
64 |
+
edge_links=pose_data.edge_links,
|
65 |
+
edge_colors=pose_data.edge_colors,
|
66 |
+
keypoint_colors=pose_data.keypoint_colors,
|
67 |
+
joint_thickness=2,
|
68 |
+
box_thickness=2,
|
69 |
+
keypoint_radius=5
|
70 |
+
)
|
71 |
+
|
72 |
+
# Convert the resulting visualization to a PIL Image
|
73 |
+
# output_image_pil = Image.fromarray(output_image.astype('uint8'), 'RGB')
|
74 |
+
|
75 |
+
# Return the PIL Image directly
|
76 |
+
return output_image, skeleton_image
|
77 |
+
|
78 |
+
# Define the Gradio interface
|
79 |
+
iface = gr.Interface(
|
80 |
+
fn=process_and_predict,
|
81 |
+
inputs=[
|
82 |
+
gr.Textbox(placeholder="Enter Image URL", label="Image URL"),
|
83 |
+
gr.Image(label="Upload Image", type='numpy'),
|
84 |
+
gr.Slider(minimum=0, maximum=1, step=0.01, value=0.5, label="Confidence Threshold"),
|
85 |
+
gr.Slider(minimum=0, maximum=1, step=0.01, value=0.5, label="IoU Threshold")
|
86 |
+
],
|
87 |
+
outputs=[
|
88 |
+
gr.components.Image(label="Estimated Pose"),
|
89 |
+
gr.components.Image(label="Skeleton Only")
|
90 |
+
],
|
91 |
+
title="YOLO-NAS-Pose Demo",
|
92 |
+
description="Upload an image, enter an image URL, or use your webcam to use a pretrained YOLO-NAS-Pose L for inference. You can check out the ",
|
93 |
+
live=False,
|
94 |
+
allow_flagging=False,
|
95 |
+
|
96 |
+
)
|
97 |
+
|
98 |
+
# Launch the interface
|
99 |
+
iface.launch()
|
100 |
+
|