Spaces:
Runtime error
Runtime error
Upload 4 files
Browse files- app.py +95 -38
- example2.jpg +0 -0
- example3.jpg +0 -0
app.py
CHANGED
@@ -6,14 +6,13 @@ from ultralytics import YOLO
|
|
6 |
from google.cloud import vision
|
7 |
_api_key = os.environ["API_KEY"]
|
8 |
_project_id = os.environ["PROJECT_ID"]
|
9 |
-
|
10 |
client = vision.ImageAnnotatorClient(client_options={"quota_project_id": _project_id, "api_key": _api_key})
|
11 |
|
12 |
import math
|
13 |
from scipy.spatial import KDTree
|
14 |
import io
|
15 |
from time import time
|
16 |
-
from PIL import Image, ImageDraw
|
17 |
import numpy as np
|
18 |
import cv2
|
19 |
|
@@ -23,9 +22,10 @@ modelPh = r'corners-best.pt'
|
|
23 |
|
24 |
model1DIM = 640
|
25 |
keypointModel = r'keypoints-best.pt'
|
|
|
26 |
|
27 |
|
28 |
-
_examples = ["example1.jpg", "example2.jpg"]
|
29 |
|
30 |
|
31 |
def unwarp_image(warped_image, src_points, dst_points, output_width, output_height):
|
@@ -57,14 +57,18 @@ def get_corners(results:list, img):
|
|
57 |
# keypoints ie corners for homography
|
58 |
KP = "topLeft topRight bottomRight bottomLeft".split()
|
59 |
r = results[0]
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
68 |
|
69 |
|
70 |
model = None
|
@@ -88,6 +92,8 @@ def preprocessImg(planar):
|
|
88 |
elif w!=h:
|
89 |
img = img.resize((_,_))
|
90 |
|
|
|
|
|
91 |
return img
|
92 |
|
93 |
|
@@ -174,7 +180,8 @@ def result_as_validvalue(contents:list[dict])->tuple[list[dict], list[str]]:
|
|
174 |
continue
|
175 |
b = f["boxCorners"]
|
176 |
m = median_point_of_bounding_box(*np.array(b).flatten())
|
177 |
-
|
|
|
178 |
|
179 |
valid.sort(key=lambda e: e["value"])
|
180 |
return valid, other
|
@@ -200,9 +207,9 @@ def determine_ocr_neighbors(center, valid:list[dict])->tuple[ list, float ]:
|
|
200 |
a = np.array(values[-1]["mid"]) - center
|
201 |
b = np.array(v["mid"]) - center
|
202 |
ang,_ = cosangle(a,b)
|
203 |
-
if _ <0:
|
204 |
-
|
205 |
-
|
206 |
angS += ang
|
207 |
u["dang"] = ang
|
208 |
u["dvda"] = u["dv"] / ang
|
@@ -250,6 +257,17 @@ def angles_from_tip(keypoints, values, nearestIx):
|
|
250 |
return values
|
251 |
|
252 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
253 |
def get_needle_value(img, keypoints):
|
254 |
|
255 |
tic2 = time()
|
@@ -260,18 +278,32 @@ def get_needle_value(img, keypoints):
|
|
260 |
assert len(contents)
|
261 |
valid,other = result_as_validvalue(contents)
|
262 |
assert len(valid)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
263 |
center = np.array(keypoints["center"])
|
264 |
values, rate = determine_ocr_neighbors(center, valid)
|
265 |
assert len(values)>=2, "failed to find at least 2 OCR values"
|
266 |
|
267 |
# import pandas as pd
|
268 |
# print(pd.DataFrame.from_dict(values))
|
269 |
-
|
270 |
-
tree = KDTree([v["mid"] for v in values])
|
271 |
-
# find bounding ocr values of tip
|
272 |
-
dist, nearestIx = tree.query(keypoints["tip"],k=2)
|
273 |
-
nearestIx.sort()
|
274 |
-
dist, nearestIx
|
275 |
|
276 |
values = angles_from_tip(keypoints, values, nearestIx)
|
277 |
# compare against start and end
|
@@ -336,36 +368,61 @@ def predict(img, detect_gauge_first):
|
|
336 |
if detect_gauge_first:
|
337 |
model0 = get_load_PhModel()
|
338 |
results = model0.predict(img)
|
339 |
-
|
|
|
|
|
340 |
else:
|
341 |
-
|
342 |
|
343 |
-
|
344 |
-
phimg
|
345 |
-
|
346 |
-
|
|
|
|
|
347 |
|
348 |
-
|
349 |
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
|
|
|
|
355 |
|
356 |
-
return
|
357 |
|
358 |
|
359 |
def test(img, detect_gauge_first):
|
360 |
return {"msg":str(img.size), "other": detect_gauge_first}
|
361 |
|
362 |
-
|
363 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
364 |
inputs=[
|
365 |
-
gr.Image(type="pil", sources=["upload"
|
366 |
gr.Checkbox(True, label="detect gauge first", info="if input image is zoomed in on only one gauge, uncheck box")
|
367 |
],
|
368 |
outputs="json",
|
369 |
-
examples=
|
370 |
cache_examples=True)\
|
371 |
.launch()
|
|
|
6 |
from google.cloud import vision
|
7 |
_api_key = os.environ["API_KEY"]
|
8 |
_project_id = os.environ["PROJECT_ID"]
|
|
|
9 |
client = vision.ImageAnnotatorClient(client_options={"quota_project_id": _project_id, "api_key": _api_key})
|
10 |
|
11 |
import math
|
12 |
from scipy.spatial import KDTree
|
13 |
import io
|
14 |
from time import time
|
15 |
+
from PIL import Image, ImageDraw, ImageFilter
|
16 |
import numpy as np
|
17 |
import cv2
|
18 |
|
|
|
22 |
|
23 |
model1DIM = 640
|
24 |
keypointModel = r'keypoints-best.pt'
|
25 |
+
minSz = 1280
|
26 |
|
27 |
|
28 |
+
_examples = [["example1.jpg",True], ["example2.jpg",False], ["example3.jpg",True]]
|
29 |
|
30 |
|
31 |
def unwarp_image(warped_image, src_points, dst_points, output_width, output_height):
|
|
|
57 |
# keypoints ie corners for homography
|
58 |
KP = "topLeft topRight bottomRight bottomLeft".split()
|
59 |
r = results[0]
|
60 |
+
planars = []
|
61 |
+
kps = []
|
62 |
+
for kpco in r.keypoints.xy.cpu():#.squeeze()
|
63 |
+
assert len(kpco)>0, "not found"
|
64 |
+
keypoints = {k:v.numpy() for v,k in zip(kpco,KP)}
|
65 |
+
sz = model1DIM
|
66 |
+
dstCorners = np.array([(0,0),(sz,0),(sz,sz),(0,sz)])
|
67 |
+
planar = unwarp_image(img, np.array(list(keypoints.values())),dstCorners, sz,sz)
|
68 |
+
# planar.save("temp-ph.jpg")
|
69 |
+
planars.append(planar)
|
70 |
+
kps.append(keypoints)
|
71 |
+
return planars, kps
|
72 |
|
73 |
|
74 |
model = None
|
|
|
92 |
elif w!=h:
|
93 |
img = img.resize((_,_))
|
94 |
|
95 |
+
if _ < minSz:
|
96 |
+
img = img.resize((minSz,minSz))
|
97 |
return img
|
98 |
|
99 |
|
|
|
180 |
continue
|
181 |
b = f["boxCorners"]
|
182 |
m = median_point_of_bounding_box(*np.array(b).flatten())
|
183 |
+
a = cv2.contourArea(np.array(b)) / len(f["text"])
|
184 |
+
valid.append({"text":f["text"], "value": value, "mid": m, "apchar":a, "box":b})
|
185 |
|
186 |
valid.sort(key=lambda e: e["value"])
|
187 |
return valid, other
|
|
|
207 |
a = np.array(values[-1]["mid"]) - center
|
208 |
b = np.array(v["mid"]) - center
|
209 |
ang,_ = cosangle(a,b)
|
210 |
+
# if _ <0:
|
211 |
+
# Warning(f"skipping {u['value']} rot:{_}")
|
212 |
+
# continue
|
213 |
angS += ang
|
214 |
u["dang"] = ang
|
215 |
u["dvda"] = u["dv"] / ang
|
|
|
257 |
return values
|
258 |
|
259 |
|
260 |
+
def sort_clockwise_with_start(coordinates, x_center, y_center, starting_index):
|
261 |
+
angles = [math.atan2(y - y_center, x - x_center) for x, y in coordinates]
|
262 |
+
sorted_indices = sorted(range(len(angles)), key=lambda i: (angles[i] - angles[starting_index] + 2 * math.pi) % (2 * math.pi))
|
263 |
+
return sorted_indices, angles
|
264 |
+
|
265 |
+
def remove_nonrange_value(valid):
|
266 |
+
meanArea = np.mean([e["apchar"] for e in valid])
|
267 |
+
cutoff = 0.5
|
268 |
+
valid = list(filter(lambda e: abs(e["apchar"]-meanArea)/meanArea < cutoff, valid))
|
269 |
+
return valid
|
270 |
+
|
271 |
def get_needle_value(img, keypoints):
|
272 |
|
273 |
tic2 = time()
|
|
|
278 |
assert len(contents)
|
279 |
valid,other = result_as_validvalue(contents)
|
280 |
assert len(valid)
|
281 |
+
|
282 |
+
valid.append({"text":"tip", "mid":keypoints["tip"]})
|
283 |
+
ix,an = sort_clockwise_with_start([e["mid"] for e in valid],*keypoints["center"], 0)
|
284 |
+
valid = [valid[i] for i in ix]
|
285 |
+
assert valid[-1]["text"]!="tip" and valid[0]["text"]!="tip", "failed to properly detect tip"
|
286 |
+
nearestIx=[]
|
287 |
+
for i,v in enumerate(valid):
|
288 |
+
if "tip"==v["text"]:
|
289 |
+
nearestIx = [i-1,i]
|
290 |
+
valid.pop(i)
|
291 |
+
break
|
292 |
+
nearestIx = np.array(nearestIx)
|
293 |
+
valid = remove_nonrange_value(valid)
|
294 |
+
|
295 |
center = np.array(keypoints["center"])
|
296 |
values, rate = determine_ocr_neighbors(center, valid)
|
297 |
assert len(values)>=2, "failed to find at least 2 OCR values"
|
298 |
|
299 |
# import pandas as pd
|
300 |
# print(pd.DataFrame.from_dict(values))
|
301 |
+
|
302 |
+
# tree = KDTree([v["mid"] for v in values])
|
303 |
+
# # find bounding ocr values of tip
|
304 |
+
# dist, nearestIx = tree.query(keypoints["tip"],k=2)
|
305 |
+
# nearestIx.sort()
|
306 |
+
# dist, nearestIx
|
307 |
|
308 |
values = angles_from_tip(keypoints, values, nearestIx)
|
309 |
# compare against start and end
|
|
|
368 |
if detect_gauge_first:
|
369 |
model0 = get_load_PhModel()
|
370 |
results = model0.predict(img)
|
371 |
+
phimgs,_ = get_corners(results, img)
|
372 |
+
if len(phimgs)==0:
|
373 |
+
raise gr.Error("no gauge found")
|
374 |
else:
|
375 |
+
phimgs = [img.copy()]
|
376 |
|
377 |
+
payloads = []
|
378 |
+
for phimg in phimgs:
|
379 |
+
model = get_load_KpModel()
|
380 |
+
phimg = preprocessImg(phimg)
|
381 |
+
results = model.predict(phimg)
|
382 |
+
keypoints = get_keypoints(results)
|
383 |
|
384 |
+
angle2tip, totalAngle = calculate_sweep_angles(keypoints)
|
385 |
|
386 |
+
phimg = phimg.filter(ImageFilter.UnsharpMask(radius=3))
|
387 |
+
payload = get_needle_value(phimg, keypoints)
|
388 |
+
payload["angleToTip"] = round(float(angle2tip),2)
|
389 |
+
payload["totalAngle"] = round(float(totalAngle),2)
|
390 |
+
for k,v in payload.items():
|
391 |
+
print(k, type(v),v)
|
392 |
+
payloads.append(payload)
|
393 |
|
394 |
+
return payloads
|
395 |
|
396 |
|
397 |
def test(img, detect_gauge_first):
|
398 |
return {"msg":str(img.size), "other": detect_gauge_first}
|
399 |
|
400 |
+
description = r"""
|
401 |
+
<b>Official 🤗 Gradio demo</b> for <a href='https://synanthropic.com/reading-analog-gauge' target='_blank'><b>Reading Analog Gauges: Automate Gauge Readings with AI in Days, Not Months
|
402 |
+
</b></a>.<br>
|
403 |
+
<br>
|
404 |
+
This model reads analog dial gauge by detecting, applying perspective correction, and gauge reading.
|
405 |
+
<br>
|
406 |
+
The model was build <i><strong>only</strong></i> with synthetic data.<br>
|
407 |
+
<br>
|
408 |
+
You can read more about it [here](https://synanthropic.com/reading-analog-gauge).
|
409 |
+
<br>
|
410 |
+
<br>
|
411 |
+
❗️Usage steps:<br>
|
412 |
+
1️⃣ Upload an image with analog dial gauge with readable values. The gauge face in the uploaded image should <b>occupy the majority of the image</b>.<br>
|
413 |
+
2️⃣ If the image has only one gauge and is a direct flat view, uncheck <strong>detect gauge first</strong>.</br>
|
414 |
+
3️⃣ Click the <b>Submit</b> button to start inference.<br>
|
415 |
+
<br>
|
416 |
+
"""
|
417 |
+
|
418 |
+
gr.Interface(title="title",
|
419 |
+
description=description,
|
420 |
+
fn=predict,
|
421 |
inputs=[
|
422 |
+
gr.Image(type="pil", sources=["upload"],streaming=False, min_width=640),
|
423 |
gr.Checkbox(True, label="detect gauge first", info="if input image is zoomed in on only one gauge, uncheck box")
|
424 |
],
|
425 |
outputs="json",
|
426 |
+
examples=_examples,
|
427 |
cache_examples=True)\
|
428 |
.launch()
|
example2.jpg
CHANGED
example3.jpg
ADDED