IbrahimHasani
commited on
Commit
•
472aaf0
1
Parent(s):
de8e664
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,609 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import requests
|
3 |
+
|
4 |
+
from PIL import Image
|
5 |
+
import PIL
|
6 |
+
from PIL import ImageDraw
|
7 |
+
|
8 |
+
from matplotlib import pyplot as plt
|
9 |
+
import matplotlib
|
10 |
+
from matplotlib import rcParams
|
11 |
+
|
12 |
+
import os
|
13 |
+
import tempfile
|
14 |
+
from io import BytesIO
|
15 |
+
from pathlib import Path
|
16 |
+
import argparse
|
17 |
+
import random
|
18 |
+
import numpy as np
|
19 |
+
import torch
|
20 |
+
import matplotlib.cm as cm
|
21 |
+
import pandas as pd
|
22 |
+
|
23 |
+
|
24 |
+
from transformers import OwlViTProcessor, OwlViTForObjectDetection
|
25 |
+
from transformers.image_utils import ImageFeatureExtractionMixin
|
26 |
+
|
27 |
+
|
28 |
+
from SuperGluePretrainedNetwork.models.matching import Matching
|
29 |
+
from SuperGluePretrainedNetwork.models.utils import (compute_pose_error, compute_epipolar_error,
|
30 |
+
estimate_pose,
|
31 |
+
error_colormap, AverageTimer, pose_auc, read_image,
|
32 |
+
rotate_intrinsics, rotate_pose_inplane,
|
33 |
+
scale_intrinsics)
|
34 |
+
|
35 |
+
torch.set_grad_enabled(False)
|
36 |
+
|
37 |
+
|
38 |
+
|
39 |
+
|
40 |
+
mixin = ImageFeatureExtractionMixin()
|
41 |
+
model = OwlViTForObjectDetection.from_pretrained("google/owlvit-base-patch32")
|
42 |
+
processor = OwlViTProcessor.from_pretrained("google/owlvit-base-patch32")
|
43 |
+
|
44 |
+
|
45 |
+
# Use GPU if available
|
46 |
+
if torch.cuda.is_available():
|
47 |
+
device = torch.device("cuda")
|
48 |
+
else:
|
49 |
+
device = torch.device("cpu")
|
50 |
+
|
51 |
+
|
52 |
+
import requests
|
53 |
+
from PIL import Image, ImageDraw
|
54 |
+
from io import BytesIO
|
55 |
+
import matplotlib.pyplot as plt
|
56 |
+
import numpy as np
|
57 |
+
import torch
|
58 |
+
import cv2
|
59 |
+
import tempfile
|
60 |
+
|
61 |
+
def detect_and_crop2(target_image_path,
|
62 |
+
query_image_path,
|
63 |
+
model,
|
64 |
+
processor,
|
65 |
+
mixin,
|
66 |
+
device,
|
67 |
+
threshold=0.5,
|
68 |
+
nms_threshold=0.3,
|
69 |
+
visualize=True):
|
70 |
+
|
71 |
+
# Open target image
|
72 |
+
image = Image.open(target_image_path).convert('RGB')
|
73 |
+
image_size = model.config.vision_config.image_size + 5
|
74 |
+
image = mixin.resize(image, image_size)
|
75 |
+
target_sizes = torch.Tensor([image.size[::-1]])
|
76 |
+
|
77 |
+
# Open query image
|
78 |
+
query_image = Image.open(query_image_path).convert('RGB')
|
79 |
+
image_size = model.config.vision_config.image_size + 5
|
80 |
+
query_image = mixin.resize(query_image, image_size)
|
81 |
+
|
82 |
+
# Process input and query image
|
83 |
+
inputs = processor(images=image, query_images=query_image, return_tensors="pt").to(device)
|
84 |
+
|
85 |
+
# Get predictions
|
86 |
+
with torch.no_grad():
|
87 |
+
outputs = model.image_guided_detection(**inputs)
|
88 |
+
|
89 |
+
# Convert predictions to CPU
|
90 |
+
img = cv2.cvtColor(np.array(image), cv2.COLOR_BGR2RGB)
|
91 |
+
outputs.logits = outputs.logits.cpu()
|
92 |
+
outputs.target_pred_boxes = outputs.target_pred_boxes.cpu()
|
93 |
+
|
94 |
+
# Post process the predictions
|
95 |
+
results = processor.post_process_image_guided_detection(outputs=outputs, threshold=threshold, nms_threshold=nms_threshold, target_sizes=target_sizes)
|
96 |
+
boxes, scores = results[0]["boxes"], results[0]["scores"]
|
97 |
+
|
98 |
+
# If no boxes, return an empty list
|
99 |
+
if len(boxes) == 0 and visualize:
|
100 |
+
print(f"No boxes detected for image: {target_image_path}")
|
101 |
+
fig, ax = plt.subplots(figsize=(6, 6))
|
102 |
+
ax.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
103 |
+
ax.set_title("Original Image")
|
104 |
+
ax.axis("off")
|
105 |
+
plt.show()
|
106 |
+
return []
|
107 |
+
|
108 |
+
# Filter boxes
|
109 |
+
img_with_all_boxes = img.copy()
|
110 |
+
filtered_boxes = []
|
111 |
+
filtered_scores = []
|
112 |
+
img_width, img_height = img.shape[1], img.shape[0]
|
113 |
+
for box, score in zip(boxes, scores):
|
114 |
+
x1, y1, x2, y2 = [int(i) for i in box.tolist()]
|
115 |
+
if x1 < 0 or y1 < 0 or x2 < 0 or y2 < 0:
|
116 |
+
continue
|
117 |
+
if (x2 - x1) / img_width >= 0.94 and (y2 - y1) / img_height >= 0.94:
|
118 |
+
continue
|
119 |
+
filtered_boxes.append([x1, y1, x2, y2])
|
120 |
+
filtered_scores.append(score)
|
121 |
+
|
122 |
+
# Draw boxes on original image
|
123 |
+
draw = ImageDraw.Draw(image)
|
124 |
+
for box in filtered_boxes:
|
125 |
+
draw.rectangle(box, outline="red",width=3)
|
126 |
+
|
127 |
+
cropped_images = []
|
128 |
+
for box in filtered_boxes:
|
129 |
+
x1, y1, x2, y2 = box
|
130 |
+
cropped_img = img[y1:y2, x1:x2]
|
131 |
+
if cropped_img.size != 0:
|
132 |
+
cropped_images.append(cropped_img)
|
133 |
+
|
134 |
+
if visualize:
|
135 |
+
# Visualization
|
136 |
+
if not filtered_boxes:
|
137 |
+
fig, ax = plt.subplots(figsize=(6, 6))
|
138 |
+
ax.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
139 |
+
ax.set_title("Original Image")
|
140 |
+
ax.axis("off")
|
141 |
+
plt.show()
|
142 |
+
else:
|
143 |
+
fig, axs = plt.subplots(1, len(cropped_images) + 2, figsize=(15, 5))
|
144 |
+
axs[0].imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
145 |
+
axs[0].set_title("Original Image")
|
146 |
+
axs[0].axis("off")
|
147 |
+
|
148 |
+
for i, (box, score) in enumerate(zip(filtered_boxes, filtered_scores)):
|
149 |
+
x1, y1, x2, y2 = box
|
150 |
+
cropped_img = img[y1:y2, x1:x2]
|
151 |
+
font = cv2.FONT_HERSHEY_SIMPLEX
|
152 |
+
text = f"{score:.2f}"
|
153 |
+
cv2.putText(cropped_img, text, (5, cropped_img.shape[0]-10), font, 0.5, (255,0,0), 1, cv2.LINE_AA)
|
154 |
+
axs[i+2].imshow(cv2.cvtColor(cropped_img, cv2.COLOR_BGR2RGB))
|
155 |
+
axs[i+2].set_title("Score: " + text)
|
156 |
+
axs[i+2].axis("off")
|
157 |
+
plt.tight_layout()
|
158 |
+
plt.show()
|
159 |
+
|
160 |
+
return cropped_images, image # return original image with boxes drawn
|
161 |
+
|
162 |
+
def save_array_to_temp_image(arr):
|
163 |
+
# Convert the array to an image
|
164 |
+
img = Image.fromarray(arr)
|
165 |
+
|
166 |
+
# Create a temporary file for the image
|
167 |
+
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.png', dir=tempfile.gettempdir())
|
168 |
+
temp_file_name = temp_file.name
|
169 |
+
temp_file.close() # We close it because we're not writing to it directly, PIL will handle the writing
|
170 |
+
|
171 |
+
# Save the image to the temp file
|
172 |
+
img.save(temp_file_name)
|
173 |
+
|
174 |
+
return temp_file_name
|
175 |
+
|
176 |
+
'''
|
177 |
+
def process_resize(w: int, h: int, resize_dims: list) -> tuple:
|
178 |
+
if len(resize_dims) == 1 and resize_dims[0] > -1:
|
179 |
+
scale = resize_dims[0] / max(h, w)
|
180 |
+
w_new, h_new = int(round(w * scale)), int(round(h * scale))
|
181 |
+
return w_new, h_new
|
182 |
+
return w, h
|
183 |
+
'''
|
184 |
+
|
185 |
+
def plot_image_pair(imgs, dpi=100, size=6, pad=.5):
|
186 |
+
n = len(imgs)
|
187 |
+
assert n == 2, 'number of images must be two'
|
188 |
+
figsize = (size*n, size*3/4) if size is not None else None
|
189 |
+
_, ax = plt.subplots(1, n, figsize=figsize, dpi=dpi)
|
190 |
+
for i in range(n):
|
191 |
+
ax[i].imshow(imgs[i], cmap=plt.get_cmap('gray'), vmin=0, vmax=255)
|
192 |
+
ax[i].get_yaxis().set_ticks([])
|
193 |
+
ax[i].get_xaxis().set_ticks([])
|
194 |
+
for spine in ax[i].spines.values(): # remove frame
|
195 |
+
spine.set_visible(False)
|
196 |
+
plt.tight_layout(pad=pad)
|
197 |
+
|
198 |
+
def plot_keypoints(kpts0, kpts1, color='w', ps=2):
|
199 |
+
ax = plt.gcf().axes
|
200 |
+
ax[0].scatter(kpts0[:, 0], kpts0[:, 1], c=color, s=ps)
|
201 |
+
ax[1].scatter(kpts1[:, 0], kpts1[:, 1], c=color, s=ps)
|
202 |
+
|
203 |
+
def plot_matches(kpts0, kpts1, color, lw=1.5, ps=4):
|
204 |
+
fig = plt.gcf()
|
205 |
+
ax = fig.axes
|
206 |
+
fig.canvas.draw()
|
207 |
+
|
208 |
+
transFigure = fig.transFigure.inverted()
|
209 |
+
fkpts0 = transFigure.transform(ax[0].transData.transform(kpts0))
|
210 |
+
fkpts1 = transFigure.transform(ax[1].transData.transform(kpts1))
|
211 |
+
|
212 |
+
fig.lines = [matplotlib.lines.Line2D(
|
213 |
+
(fkpts0[i, 0], fkpts1[i, 0]), (fkpts0[i, 1], fkpts1[i, 1]), zorder=1,
|
214 |
+
transform=fig.transFigure, c=color[i], linewidth=lw)
|
215 |
+
for i in range(len(kpts0))]
|
216 |
+
ax[0].scatter(kpts0[:, 0], kpts0[:, 1], c=color, s=ps)
|
217 |
+
ax[1].scatter(kpts1[:, 0], kpts1[:, 1], c=color, s=ps)
|
218 |
+
|
219 |
+
def unified_matching_plot2(image0, image1, kpts0, kpts1, mkpts0, mkpts1,
|
220 |
+
color, text, path=None, show_keypoints=False,
|
221 |
+
fast_viz=False, opencv_display=False,
|
222 |
+
opencv_title='matches', small_text=[]):
|
223 |
+
|
224 |
+
# Set the background color for the plot
|
225 |
+
plt.figure(facecolor='#eeeeee')
|
226 |
+
plot_image_pair([image0, image1])
|
227 |
+
|
228 |
+
# Elegant points and lines for matches
|
229 |
+
if show_keypoints:
|
230 |
+
plot_keypoints(kpts0, kpts1, color='k', ps=4)
|
231 |
+
plot_keypoints(kpts0, kpts1, color='w', ps=2)
|
232 |
+
plot_matches(mkpts0, mkpts1, color, lw=1)
|
233 |
+
|
234 |
+
fig = plt.gcf()
|
235 |
+
|
236 |
+
# Add text
|
237 |
+
fig.text(
|
238 |
+
0.01, 0.01, '\n'.join(small_text), transform=fig.axes[0].transAxes,
|
239 |
+
fontsize=10, va='bottom', ha='left', color='#333333', fontweight='bold', fontname='Helvetica',
|
240 |
+
bbox=dict(facecolor='white', alpha=0.7, edgecolor='none', boxstyle="round,pad=0.3"))
|
241 |
+
|
242 |
+
fig.text(
|
243 |
+
0.01, 0.99, '\n'.join(text), transform=fig.axes[0].transAxes,
|
244 |
+
fontsize=15, va='top', ha='left', color='#333333', fontweight='bold', fontname='Helvetica',
|
245 |
+
bbox=dict(facecolor='white', alpha=0.7, edgecolor='none', boxstyle="round,pad=0.3"))
|
246 |
+
|
247 |
+
# Optional: remove axis for a cleaner look
|
248 |
+
plt.axis('off')
|
249 |
+
|
250 |
+
# Convert the figure to an OpenCV image
|
251 |
+
buf = BytesIO()
|
252 |
+
plt.savefig(buf, format='png', bbox_inches='tight', pad_inches=0)
|
253 |
+
buf.seek(0)
|
254 |
+
img_arr = np.frombuffer(buf.getvalue(), dtype=np.uint8)
|
255 |
+
buf.close()
|
256 |
+
img = cv2.imdecode(img_arr, 1)
|
257 |
+
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
|
258 |
+
|
259 |
+
# Close the figure to free memory
|
260 |
+
plt.close(fig)
|
261 |
+
|
262 |
+
return img
|
263 |
+
|
264 |
+
def create_image_pyramid2(image_path, longest_side, scales=[0.25, 0.5, 1.0]):
|
265 |
+
original_image = cv2.imread(image_path)
|
266 |
+
oh, ow, _ = original_image.shape
|
267 |
+
|
268 |
+
# Determine the scaling factor based on the longest side
|
269 |
+
if oh > ow:
|
270 |
+
output_height = longest_side
|
271 |
+
output_width = int((ow / oh) * longest_side)
|
272 |
+
else:
|
273 |
+
output_width = longest_side
|
274 |
+
output_height = int((oh / ow) * longest_side)
|
275 |
+
output_size = (output_width, output_height)
|
276 |
+
|
277 |
+
pyramid = []
|
278 |
+
|
279 |
+
for scale in scales:
|
280 |
+
# Resize based on the scale factor
|
281 |
+
resized = cv2.resize(original_image, None, fx=scale, fy=scale)
|
282 |
+
rh, rw, _ = resized.shape
|
283 |
+
|
284 |
+
if scale < 1.0: # downsampling
|
285 |
+
# Calculate the amount of padding required
|
286 |
+
dy_top = max((output_size[1] - rh) // 2, 0)
|
287 |
+
dy_bottom = output_size[1] - rh - dy_top
|
288 |
+
dx_left = max((output_size[0] - rw) // 2, 0)
|
289 |
+
dx_right = output_size[0] - rw - dx_left
|
290 |
+
|
291 |
+
# Create padded image
|
292 |
+
padded = cv2.copyMakeBorder(resized, dy_top, dy_bottom, dx_left, dx_right, cv2.BORDER_CONSTANT, value=[255, 255, 255])
|
293 |
+
pyramid.append(padded)
|
294 |
+
elif scale > 1.0: # upsampling
|
295 |
+
# We need to crop the image to fit the desired output size
|
296 |
+
dy = (rh - output_size[1]) // 2
|
297 |
+
dx = (rw - output_size[0]) // 2
|
298 |
+
cropped = resized[dy:dy+output_size[1], dx:dx+output_size[0]]
|
299 |
+
pyramid.append(cropped)
|
300 |
+
else: # scale == 1.0
|
301 |
+
pyramid.append(resized)
|
302 |
+
|
303 |
+
return pyramid
|
304 |
+
|
305 |
+
# Example usage
|
306 |
+
# pyramid = create_image_pyramid('path_to_image.jpg', 800)
|
307 |
+
def image_matching(query_img, target_img, image_dims=[640*2], scale_factors=[0.33,0.66,1], visualize=True, k_thresh=None, m_thresh=None, write=False):
|
308 |
+
|
309 |
+
image1, inp1, scales1 = read_image(target_img, device, [640*2], 0, True)
|
310 |
+
query_pyramid = create_image_pyramid2(query_img, image_dims[0], scale_factors)
|
311 |
+
|
312 |
+
all_valid = []
|
313 |
+
all_inliers = []
|
314 |
+
all_return_imgs = []
|
315 |
+
max_matches_img = None
|
316 |
+
max_matches = -1
|
317 |
+
|
318 |
+
for idx, query_level in enumerate(query_pyramid):
|
319 |
+
temp_file_path = "temp_level_{}.png".format(idx)
|
320 |
+
cv2.imwrite(temp_file_path, query_level)
|
321 |
+
|
322 |
+
image0, inp0, scales0 = read_image(temp_file_path, device, [640*2], 0, True)
|
323 |
+
|
324 |
+
if image0 is None or image1 is None:
|
325 |
+
print('Problem reading image pair: {} {}'.format(query_img, target_img))
|
326 |
+
else:
|
327 |
+
# Matching
|
328 |
+
pred = matching({'image0': inp0, 'image1': inp1})
|
329 |
+
pred = {k: v[0] for k, v in pred.items()}
|
330 |
+
kpts0, kpts1 = pred['keypoints0'], pred['keypoints1']
|
331 |
+
matches, conf = pred['matches0'], pred['matching_scores0']
|
332 |
+
|
333 |
+
valid = matches > -1
|
334 |
+
mkpts0 = kpts0[valid]
|
335 |
+
mkpts1 = kpts1[matches[valid]]
|
336 |
+
mconf = conf[valid]
|
337 |
+
#color = cm.jet(mconf)[:len(mkpts0)] # Ensure consistent size
|
338 |
+
color = cm.jet(mconf.detach().numpy())[:len(mkpts0)]
|
339 |
+
|
340 |
+
all_valid.append(np.sum( valid.tolist() ))
|
341 |
+
|
342 |
+
# Convert torch tensors to numpy arrays.
|
343 |
+
mkpts0_np = mkpts0.cpu().numpy()
|
344 |
+
mkpts1_np = mkpts1.cpu().numpy()
|
345 |
+
|
346 |
+
try:
|
347 |
+
# Use RANSAC to find the homography matrix.
|
348 |
+
H, inliers = cv2.findHomography(mkpts0_np, mkpts1_np, cv2.RANSAC, 5.0)
|
349 |
+
except:
|
350 |
+
H = 0
|
351 |
+
inliers = 0
|
352 |
+
print ("Not enough points for homography")
|
353 |
+
# Convert inliers from shape (N, 1) to shape (N,) and count them.
|
354 |
+
num_inliers = np.sum(inliers)
|
355 |
+
|
356 |
+
all_inliers.append(num_inliers)
|
357 |
+
|
358 |
+
# Visualization
|
359 |
+
text = [
|
360 |
+
'Engagify Image Matching',
|
361 |
+
'Keypoints: {}:{}'.format(len(kpts0), len(kpts1)),
|
362 |
+
'Scaling Factor: {}'.format( scale_factors[idx]),
|
363 |
+
'Matches: {}'.format(len(mkpts0)),
|
364 |
+
'Inliers: {}'.format(num_inliers),
|
365 |
+
]
|
366 |
+
|
367 |
+
|
368 |
+
k_thresh = matching.superpoint.config['keypoint_threshold']
|
369 |
+
m_thresh = matching.superglue.config['match_threshold']
|
370 |
+
|
371 |
+
small_text = [
|
372 |
+
'Keypoint Threshold: {:.4f}'.format(k_thresh),
|
373 |
+
'Match Threshold: {:.2f}'.format(m_thresh),
|
374 |
+
]
|
375 |
+
|
376 |
+
visualized_img = None # To store the visualized image
|
377 |
+
|
378 |
+
if visualize:
|
379 |
+
ret_img = unified_matching_plot2(
|
380 |
+
image0, image1, kpts0, kpts1, mkpts0, mkpts1, color, text, 'Test_Level_{}'.format(idx), True, False, True, 'Matches_Level_{}'.format(idx), small_text)
|
381 |
+
all_return_imgs.append(ret_img)
|
382 |
+
# Storing image with most matches
|
383 |
+
#if len(mkpts0) > max_matches:
|
384 |
+
# max_matches = len(mkpts0)
|
385 |
+
# max_matches_img = 'Matches_Level_{}'.format(idx)
|
386 |
+
|
387 |
+
avg_valid = np.sum(all_valid) / len(scale_factors)
|
388 |
+
avg_inliers = np.sum(all_inliers) / len(scale_factors)
|
389 |
+
|
390 |
+
# Convert the image with the most matches to base64 encoded format
|
391 |
+
# with open(max_matches_img, "rb") as image_file:
|
392 |
+
# encoded_string = base64.b64encode(image_file.read()).decode()
|
393 |
+
|
394 |
+
return {'valid':all_valid, 'inliers':all_inliers, 'visualized_image':all_return_imgs} #, encoded_string
|
395 |
+
|
396 |
+
# Usage:
|
397 |
+
#results = image_matching('Samples/Poster/poster_event_small_22.jpg', 'Samples/Images/16.jpeg', visualize=True)
|
398 |
+
#print (results)
|
399 |
+
|
400 |
+
def image_matching_no_pyramid(query_img, target_img, visualize=True, write=False):
|
401 |
+
|
402 |
+
image1, inp1, scales1 = read_image(target_img, device, [640*2], 0, True)
|
403 |
+
image0, inp0, scales0 = read_image(query_img, device, [640*2], 0, True)
|
404 |
+
|
405 |
+
if image0 is None or image1 is None:
|
406 |
+
print('Problem reading image pair: {} {}'.format(query_img, target_img))
|
407 |
+
return None
|
408 |
+
|
409 |
+
# Matching
|
410 |
+
pred = matching({'image0': inp0, 'image1': inp1})
|
411 |
+
pred = {k: v[0] for k, v in pred.items()}
|
412 |
+
kpts0, kpts1 = pred['keypoints0'], pred['keypoints1']
|
413 |
+
matches, conf = pred['matches0'], pred['matching_scores0']
|
414 |
+
|
415 |
+
valid = matches > -1
|
416 |
+
mkpts0 = kpts0[valid]
|
417 |
+
mkpts1 = kpts1[matches[valid]]
|
418 |
+
mconf = conf[valid]
|
419 |
+
#color = cm.jet(mconf)[:len(mkpts0)] # Ensure consistent size
|
420 |
+
color = cm.jet(mconf.detach().numpy())[:len(mkpts0)]
|
421 |
+
|
422 |
+
valid_count = np.sum(valid.tolist())
|
423 |
+
|
424 |
+
# Convert torch tensors to numpy arrays.
|
425 |
+
mkpts0_np = mkpts0.cpu().numpy()
|
426 |
+
mkpts1_np = mkpts1.cpu().numpy()
|
427 |
+
|
428 |
+
try:
|
429 |
+
# Use RANSAC to find the homography matrix.
|
430 |
+
H, inliers = cv2.findHomography(mkpts0_np, mkpts1_np, cv2.RANSAC, 5.0)
|
431 |
+
except:
|
432 |
+
H = 0
|
433 |
+
inliers = 0
|
434 |
+
print("Not enough points for homography")
|
435 |
+
|
436 |
+
# Convert inliers from shape (N, 1) to shape (N,) and count them.
|
437 |
+
num_inliers = np.sum(inliers)
|
438 |
+
|
439 |
+
# Visualization
|
440 |
+
text = [
|
441 |
+
'Engagify Image Matching',
|
442 |
+
'Keypoints: {}:{}'.format(len(kpts0), len(kpts1)),
|
443 |
+
'Matches: {}'.format(len(mkpts0)),
|
444 |
+
'Inliers: {}'.format(num_inliers),
|
445 |
+
]
|
446 |
+
|
447 |
+
k_thresh = matching.superpoint.config['keypoint_threshold']
|
448 |
+
m_thresh = matching.superglue.config['match_threshold']
|
449 |
+
|
450 |
+
small_text = [
|
451 |
+
'Keypoint Threshold: {:.4f}'.format(k_thresh),
|
452 |
+
'Match Threshold: {:.2f}'.format(m_thresh),
|
453 |
+
]
|
454 |
+
|
455 |
+
visualized_img = None # To store the visualized image
|
456 |
+
|
457 |
+
if visualize:
|
458 |
+
visualized_img = unified_matching_plot2(
|
459 |
+
image0, image1, kpts0, kpts1, mkpts0, mkpts1, color, text, 'Test_Match', True, False, True, 'Matches', small_text)
|
460 |
+
|
461 |
+
return {
|
462 |
+
'valid': [valid_count],
|
463 |
+
'inliers': [num_inliers],
|
464 |
+
'visualized_image': [visualized_img]
|
465 |
+
}
|
466 |
+
|
467 |
+
# Usage:
|
468 |
+
#results = image_matching_no_pyramid('Samples/Poster/poster_event_small_22.jpg', 'Samples/Images/16.jpeg', visualize=True)
|
469 |
+
|
470 |
+
# Load the SuperPoint and SuperGlue models.
|
471 |
+
device = 'cuda' if torch.cuda.is_available() and not opt.force_cpu else 'cpu'
|
472 |
+
print('Running inference on device \"{}\"'.format(device))
|
473 |
+
config = {
|
474 |
+
'superpoint': {
|
475 |
+
'nms_radius': 4,
|
476 |
+
'keypoint_threshold': 0.005,
|
477 |
+
'max_keypoints': 1024
|
478 |
+
},
|
479 |
+
'superglue': {
|
480 |
+
'weights': 'outdoor',
|
481 |
+
'sinkhorn_iterations': 20,
|
482 |
+
'match_threshold': 0.2,
|
483 |
+
}
|
484 |
+
}
|
485 |
+
matching = Matching(config).eval().to(device)
|
486 |
+
|
487 |
+
from PIL import Image
|
488 |
+
|
489 |
+
def stitch_images(images):
|
490 |
+
"""Stitches a list of images vertically."""
|
491 |
+
if not images:
|
492 |
+
# Return a placeholder image if the images list is empty
|
493 |
+
return Image.new('RGB', (100, 100), color='gray')
|
494 |
+
|
495 |
+
max_width = max([img.width for img in images])
|
496 |
+
total_height = sum(img.height for img in images)
|
497 |
+
|
498 |
+
composite = Image.new('RGB', (max_width, total_height))
|
499 |
+
|
500 |
+
y_offset = 0
|
501 |
+
for img in images:
|
502 |
+
composite.paste(img, (0, y_offset))
|
503 |
+
y_offset += img.height
|
504 |
+
|
505 |
+
return composite
|
506 |
+
|
507 |
+
def check_object_in_image3(query_image, target_image, threshold=50, scale_factor=[0.33,0.66,1]):
|
508 |
+
decision_on = []
|
509 |
+
# Convert cv2 images to PIL images and add them to a list
|
510 |
+
images_to_return = []
|
511 |
+
|
512 |
+
cropped_images, bbox_image = detect_and_crop2(target_image_path=target_image,
|
513 |
+
query_image_path=query_image,
|
514 |
+
model=model,
|
515 |
+
processor=processor,
|
516 |
+
mixin=mixin,
|
517 |
+
device=device,
|
518 |
+
visualize=False)
|
519 |
+
|
520 |
+
temp_files = [save_array_to_temp_image(i) for i in cropped_images]
|
521 |
+
crop_results = [image_matching_no_pyramid(query_image, i, visualize=True) for i in temp_files]
|
522 |
+
|
523 |
+
cropped_visuals = []
|
524 |
+
cropped_inliers = []
|
525 |
+
for result in crop_results:
|
526 |
+
# Add visualized images to the temporary list
|
527 |
+
for img in result['visualized_image']:
|
528 |
+
cropped_visuals.append(Image.fromarray(img))
|
529 |
+
for inliers_ in result['inliers']:
|
530 |
+
cropped_inliers.append(inliers_)
|
531 |
+
# Stitch the cropped visuals into one image
|
532 |
+
images_to_return.append(stitch_images(cropped_visuals))
|
533 |
+
|
534 |
+
pyramid_results = image_matching(query_image, target_image, visualize=True, scale_factors=scale_factor)
|
535 |
+
|
536 |
+
pyramid_visuals = [Image.fromarray(img) for img in pyramid_results['visualized_image']]
|
537 |
+
# Stitch the pyramid visuals into one image
|
538 |
+
images_to_return.append(stitch_images(pyramid_visuals))
|
539 |
+
|
540 |
+
# Check inliers and determine if the object is present
|
541 |
+
print (cropped_inliers)
|
542 |
+
is_present = any(value > threshold for value in cropped_inliers)
|
543 |
+
if is_present == True:
|
544 |
+
decision_on.append('Object Detection')
|
545 |
+
is_present = any(value > threshold for value in pyramid_results["inliers"])
|
546 |
+
if is_present == True:
|
547 |
+
decision_on.append('Pyramid Max Point')
|
548 |
+
if is_present == False:
|
549 |
+
decision_on.append("Neither, It Failed All Tests")
|
550 |
+
|
551 |
+
# Return results as a dictionary
|
552 |
+
return {
|
553 |
+
'is_present': is_present,
|
554 |
+
'images': images_to_return,
|
555 |
+
'scale factors': scale_factor,
|
556 |
+
'object detection inliers': cropped_inliers,
|
557 |
+
'pyramid_inliers' : pyramid_results["inliers"],
|
558 |
+
'bbox_image':bbox_image,
|
559 |
+
'decision_on':decision_on,
|
560 |
+
|
561 |
+
}
|
562 |
+
|
563 |
+
# Example call:
|
564 |
+
#result = check_object_in_image3('Samples/Poster/poster_event_small.jpg', 'Samples/Images/True_Image_3423234.jpeg', 50)
|
565 |
+
# Accessing the results:
|
566 |
+
#print(result['is_present']) # prints True/False
|
567 |
+
#print(result['images']) # is a list of 2 stitched images.
|
568 |
+
|
569 |
+
|
570 |
+
import gradio as gr
|
571 |
+
import cv2
|
572 |
+
from PIL import Image
|
573 |
+
|
574 |
+
def gradio_interface(query_image_path, target_image_path, threshold):
|
575 |
+
result = check_object_in_image3(query_image_path, target_image_path, threshold)
|
576 |
+
# Depending on how many images are in the list, you can return them like this:
|
577 |
+
return result['bbox_image'], result['images'][0], result['object detection inliers'], result['scale factors'], result['pyramid_inliers'], result['images'][1], str(result['is_present']), result['decision_on']
|
578 |
+
|
579 |
+
|
580 |
+
# Define the Gradio interface
|
581 |
+
interface = gr.Interface(
|
582 |
+
fn=gradio_interface, # function to be called on button press
|
583 |
+
inputs=[
|
584 |
+
gr.components.Image(label="Query Image (Drop the Image you want to detect here)", type="filepath"),
|
585 |
+
gr.components.Image(label="Target Image (Drop the Image youd like to search here)", type="filepath"),
|
586 |
+
gr.components.Slider(minimum=0, maximum=200, value=50, step=5, label="Enter the Inlier Threshold"),
|
587 |
+
],
|
588 |
+
outputs=[
|
589 |
+
gr.components.Image(label='Filtered Regions of Interest (Candidates)'),
|
590 |
+
gr.components.Image(label="Cropped Visuals from Image Guided Object Detection "),
|
591 |
+
gr.components.Text(label='Inliers detected for Image Guided Object Detection '),
|
592 |
+
gr.components.Text(label='Scale Factors Used for Pyramid (Results below, In Order)'),
|
593 |
+
gr.components.Text(label='Inliers detected for Pyramid Search (In Order)'),
|
594 |
+
gr.components.Image(label="Pyramid Visuals"),
|
595 |
+
gr.components.Textbox(label="Object Present?"),
|
596 |
+
gr.components.Textbox(label="Decision Taken Based on?"),
|
597 |
+
],
|
598 |
+
theme=gr.themes.Monochrome(),
|
599 |
+
title="Engajify's Image Specific Image Recognition + Matching Tool",
|
600 |
+
description="[Author: Ibrahim Hasani] \n "
|
601 |
+
" This tool leverages Transformer, Deep Learning, and Traditional Computer Vision techniques to determine if a specified object "
|
602 |
+
"(given by the query image) is present within a target image. \n"
|
603 |
+
"1. Image-Guided Object Detection where we detect potential regions of interest. (Owl-Vit-Google). \n"
|
604 |
+
"2. Pyramid Search that looks at various scales of the target image. Results provide "
|
605 |
+
"visual representations of the matching process and a final verdict on the object's presence.\n"
|
606 |
+
"3. SuperPoint (MagicLeap) + SuperGlue + Homography to extract inliers, which are thresholded for decision making."
|
607 |
+
)
|
608 |
+
|
609 |
+
interface.launch()
|