Commit @6657e94befd3a8657cd7f4571d39824617925c07

윤영준 05-21

fixing a critical bug

@6657e94befd3a8657cd7f4571d39824617925c07

e379e60

6657e94

inference_endpoint.py

--- inference_endpoint.py

+++ inference_endpoint.py


 
         t1 = time.time()
         detections, self.mask = inference_engine.run_inference(cv2.resize(image, model_input_shape))
+
         t2 = time.time()
         if len(self.mask) > 0:
             print(self.mask.shape)
+            print(type(self.mask))
             self.mask_blob = cv2.imencode('.png', self.mask)
             self.mask_blob = self.mask.tobytes()
-
+            self.mask = cv2.resize(self.mask, (image.shape[0], image.shape[1]))
 
         # print(t2 - t1)
 

             self.area_percent = 0
         else :
             self.area_percent = np.sum(self.mask) / image.shape[0] * image.shape[1]
-
-
 
         # self.send_result()
         # write another post request for pushing a detection result

e379e60

6657e94

yoloseg/inference_.py

--- yoloseg/inference_.py

+++ yoloseg/inference_.py


             thres = self.model_score_threshold
             w_thres = 20
             h_thres = 20
+
+            x, y, w, h = detection[:4]
+            # if bboxes are too small, it just skips, and it is not a bad idea since we do not need to detect small areas
+            if w < w_thres or h < h_thres:
+                continue
+
             if confidence > thres:
-                x, y, w, h = detection[:4]
-                # if bboxes are too small, it just skips, and it is not a bad idea since we do not need to detect small areas
-                if w < w_thres or h < h_thres:
-                    continue
+
                 left = int((x - 0.5 * w) * x_factor)
                 top = int((y - 0.5 * h) * y_factor)
                 width = int(w * x_factor)

         for idx, det in enumerate(detections):
             box = det['box']
             x1, y1, w, h = box
-            print(f"x1 : {x1}, y1 : {y1}, w: {w}, h: {h}")
+            # print(f"x1 : {x1}, y1 : {y1}, w: {w}, h: {h}")
 
             x1, y1, x2, y2 = x1, y1, x1 + w, y1 + h
 

             if x2 > image_shape[0]:
                 w = image_shape[1] - y1
 
-            print(f"x2: {x2}, y2 : {y2}")
+            # print(f"x2: {x2}, y2 : {y2}")
 
             # Get the corresponding mask coefficients for this detection
             coeffs = det["mask_coefficients"]

             # Combine the mask with the masks of other detections
             full_masks[idx] = full_mask
         all_mask = full_masks.sum(axis=0)
-        return all_mask
+        # Append a dimension so that cv2 can understand this as an image.
+        all_mask = all_mask.reshape((image_shape[0], image_shape[1], 1))
+        return all_mask.astype(np.uint8)
 
     def load_classes_from_file(self):
         with open(self.classes_path, 'r') as f:

Add a comment

Open 0
Closed 0

List

...	...	@@ -79,12 +79,14 @@
79	79
80	80	t1 = time.time()
81	81	detections, self.mask = inference_engine.run_inference(cv2.resize(image, model_input_shape))
	82	+
82	83	t2 = time.time()
83	84	if len(self.mask) > 0:
84	85	print(self.mask.shape)
	86	+ print(type(self.mask))
85	87	self.mask_blob = cv2.imencode('.png', self.mask)
86	88	self.mask_blob = self.mask.tobytes()
87		-
	89	+ self.mask = cv2.resize(self.mask, (image.shape[0], image.shape[1]))
88	90
89	91	# print(t2 - t1)
90	92
...	...	@@ -93,8 +95,6 @@
93	95	self.area_percent = 0
94	96	else :
95	97	self.area_percent = np.sum(self.mask) / image.shape[0] * image.shape[1]
96		-
97		-
98	98
99	99	# self.send_result()
100	100	# write another post request for pushing a detection result

...	...	@@ -60,11 +60,14 @@
60	60	thres = self.model_score_threshold
61	61	w_thres = 20
62	62	h_thres = 20
	63	+
	64	+ x, y, w, h = detection[:4]
	65	+ # if bboxes are too small, it just skips, and it is not a bad idea since we do not need to detect small areas
	66	+ if w < w_thres or h < h_thres:
	67	+ continue
	68	+
63	69	if confidence > thres:
64		- x, y, w, h = detection[:4]
65		- # if bboxes are too small, it just skips, and it is not a bad idea since we do not need to detect small areas
66		- if w < w_thres or h < h_thres:
67		- continue
	70	+
68	71	left = int((x - 0.5 * w) * x_factor)
69	72	top = int((y - 0.5 * h) * y_factor)
70	73	width = int(w * x_factor)
...	...	@@ -102,7 +105,7 @@
102	105	for idx, det in enumerate(detections):
103	106	box = det['box']
104	107	x1, y1, w, h = box
105		- print(f"x1 : {x1}, y1 : {y1}, w: {w}, h: {h}")
	108	+ # print(f"x1 : {x1}, y1 : {y1}, w: {w}, h: {h}")
106	109
107	110	x1, y1, x2, y2 = x1, y1, x1 + w, y1 + h
108	111
...	...	@@ -120,7 +123,7 @@
120	123	if x2 > image_shape[0]:
121	124	w = image_shape[1] - y1
122	125
123		- print(f"x2: {x2}, y2 : {y2}")
	126	+ # print(f"x2: {x2}, y2 : {y2}")
124	127
125	128	# Get the corresponding mask coefficients for this detection
126	129	coeffs = det["mask_coefficients"]
...	...	@@ -149,7 +152,9 @@
149	152	# Combine the mask with the masks of other detections
150	153	full_masks[idx] = full_mask
151	154	all_mask = full_masks.sum(axis=0)
152		- return all_mask
	155	+ # Append a dimension so that cv2 can understand this as an image.
	156	+ all_mask = all_mask.reshape((image_shape[0], image_shape[1], 1))
	157	+ return all_mask.astype(np.uint8)
153	158
154	159	def load_classes_from_file(self):
155	160	with open(self.classes_path, 'r') as f:

Delete comment