Commit @9b92e6481b61d0fab11babee734d23e45844e653

윤영준 09-12

bug fix : wrong operation on segmentation masks (resize on entire proto mask?;;;)

@9b92e6481b61d0fab11babee734d23e45844e653

1a08293

9b92e64

yoloseg/inference_gpu_.py

--- yoloseg/inference_gpu_.py

+++ yoloseg/inference_gpu_.py


             # for now, plural batch operation is not supported, and this is the point where you should start.
             # instead of hardcoded proto_masks[0], do some iterative/vectorize operation
             mask = np.tensordot(coeffs, proto_masks[0], axes=[0, 0])  # Dot product along the number of prototypes
-
+            resized_mask = cv2.resize(mask,(image_shape[0], image_shape[1]))
             # Resize mask to the bounding box size, using sigmoid to normalize
-            resized_mask = cv2.resize(mask, (w, h))
-            resized_mask = self.sigmoid(resized_mask)
+            cropped_mask = resized_mask[y1:y1+h, x1:x1+w]
+            resized_mask = self.sigmoid(cropped_mask)
 
             # Threshold to create a binary mask
             final_mask = (resized_mask > 0.5).astype(np.uint8)

 
 def test():
     import time
-
+    import glob
+    import os
     # Path to your ONNX model and classes text file
     model_path = 'yoloseg/weight/best.onnx'
     classes_txt_file = 'config_files/yolo_config.txt'
     # image_path = 'yoloseg/img3.jpg'
-    image_path = 'yoloseg/img3.jpg'
+    image_path = 'yoloseg/img.jpg'
 
-    model_input_shape = (640, 640)
+    model_input_shape = (480, 480)
     inference_engine = Inference(
         onnx_model_path=model_path,
         model_input_shape=model_input_shape,

     img = cv2.resize(img, model_input_shape)
     # Run inference
 
-    for i in range(10):
+    # for i in range(10):
+    #     t1 = time.time()
+    #     detections, mask_maps = inference_engine.run_inference(img)
+    #     t2 = time.time()
+    #     print(t2 - t1)
+
+    images = glob.glob("/home/juni/사진/flood/out-/*.jpg")
+    images = sorted(images)
+    for k, image in enumerate(images):
+        image = cv2.imread(image)
         t1 = time.time()
-        detections, mask_maps = inference_engine.run_inference(img)
+        image = cv2.resize(image, model_input_shape)
+        detections, mask_maps = inference_engine.run_inference(image)
+
+        # Display results
+        for detection in detections:
+            x, y, w, h = detection['box']
+            class_name = detection['class_name']
+            confidence = detection['confidence']
+            cv2.rectangle(image, (x, y), (x+w, y+h), detection['color'], 2)
+            label = f"{class_name}: {confidence:.2f}"
+            cv2.putText(image, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, detection['color'], 2)
+        if len(mask_maps) != 0:
+            for i in range(mask_maps.shape[2]):  # Iterate over each mask
+                seg_image = overlay_mask(image, mask_maps[:, :, i], color=(0, 255, 0), alpha=0.3)
+                # cv2.imshow(f"Segmentation {i + 1}", seg_image)
+            # cv2.waitKey(0)  # Wait for a key press before showing the next mask
+            # cv2.destroyAllWindows()
         t2 = time.time()
         print(t2 - t1)
-
-
-
-    # Display results
-    for detection in detections:
-        x, y, w, h = detection['box']
-        class_name = detection['class_name']
-        confidence = detection['confidence']
-        cv2.rectangle(img, (x, y), (x+w, y+h), detection['color'], 2)
-        label = f"{class_name}: {confidence:.2f}"
-        cv2.putText(img, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, detection['color'], 2)
+        cv2.imwrite( f"/home/juni/사진/flood/infer/{k}.jpg", seg_image)
 
     # Show the image_binary
     # cv2.imshow('Detections', img)
     # cv2.waitKey(0)
     # cv2.destroyAllWindows()
 
-    # If you also want to display segmentation maps, you would need additional handling here
-    # Example for displaying first mask if available:
-    if len(mask_maps) != 0:
 
-        seg_image = overlay_mask(img, mask_maps[:,:,0], color=(0, 255, 0), alpha=0.3)
-        cv2.imshow("segmentation", seg_image)
-        cv2.waitKey(0)
-        cv2.destroyAllWindows()
 
 
 if __name__ == "__main__":

Add a comment

Open 0
Closed 0

List

...	...	@@ -166,10 +166,10 @@
166	166	# for now, plural batch operation is not supported, and this is the point where you should start.
167	167	# instead of hardcoded proto_masks[0], do some iterative/vectorize operation
168	168	mask = np.tensordot(coeffs, proto_masks[0], axes=[0, 0]) # Dot product along the number of prototypes
169		-
	169	+ resized_mask = cv2.resize(mask,(image_shape[0], image_shape[1]))
170	170	# Resize mask to the bounding box size, using sigmoid to normalize
171		- resized_mask = cv2.resize(mask, (w, h))
172		- resized_mask = self.sigmoid(resized_mask)
	171	+ cropped_mask = resized_mask[y1:y1+h, x1:x1+w]
	172	+ resized_mask = self.sigmoid(cropped_mask)
173	173
174	174	# Threshold to create a binary mask
175	175	final_mask = (resized_mask > 0.5).astype(np.uint8)
...	...	@@ -250,14 +250,15 @@
250	250
251	251	def test():
252	252	import time
253		-
	253	+ import glob
	254	+ import os
254	255	# Path to your ONNX model and classes text file
255	256	model_path = 'yoloseg/weight/best.onnx'
256	257	classes_txt_file = 'config_files/yolo_config.txt'
257	258	# image_path = 'yoloseg/img3.jpg'
258		- image_path = 'yoloseg/img3.jpg'
	259	+ image_path = 'yoloseg/img.jpg'
259	260
260		- model_input_shape = (640, 640)
	261	+ model_input_shape = (480, 480)
261	262	inference_engine = Inference(
262	263	onnx_model_path=model_path,
263	264	model_input_shape=model_input_shape,
...	...	@@ -273,36 +274,44 @@
273	274	img = cv2.resize(img, model_input_shape)
274	275	# Run inference
275	276
276		- for i in range(10):
	277	+ # for i in range(10):
	278	+ # t1 = time.time()
	279	+ # detections, mask_maps = inference_engine.run_inference(img)
	280	+ # t2 = time.time()
	281	+ # print(t2 - t1)
	282	+
	283	+ images = glob.glob("/home/juni/사진/flood/out-/*.jpg")
	284	+ images = sorted(images)
	285	+ for k, image in enumerate(images):
	286	+ image = cv2.imread(image)
277	287	t1 = time.time()
278		- detections, mask_maps = inference_engine.run_inference(img)
	288	+ image = cv2.resize(image, model_input_shape)
	289	+ detections, mask_maps = inference_engine.run_inference(image)
	290	+
	291	+ # Display results
	292	+ for detection in detections:
	293	+ x, y, w, h = detection['box']
	294	+ class_name = detection['class_name']
	295	+ confidence = detection['confidence']
	296	+ cv2.rectangle(image, (x, y), (x+w, y+h), detection['color'], 2)
	297	+ label = f"{class_name}: {confidence:.2f}"
	298	+ cv2.putText(image, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, detection['color'], 2)
	299	+ if len(mask_maps) != 0:
	300	+ for i in range(mask_maps.shape[2]): # Iterate over each mask
	301	+ seg_image = overlay_mask(image, mask_maps[:, :, i], color=(0, 255, 0), alpha=0.3)
	302	+ # cv2.imshow(f"Segmentation {i + 1}", seg_image)
	303	+ # cv2.waitKey(0) # Wait for a key press before showing the next mask
	304	+ # cv2.destroyAllWindows()
279	305	t2 = time.time()
280	306	print(t2 - t1)
281		-
282		-
283		-
284		- # Display results
285		- for detection in detections:
286		- x, y, w, h = detection['box']
287		- class_name = detection['class_name']
288		- confidence = detection['confidence']
289		- cv2.rectangle(img, (x, y), (x+w, y+h), detection['color'], 2)
290		- label = f"{class_name}: {confidence:.2f}"
291		- cv2.putText(img, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, detection['color'], 2)
	307	+ cv2.imwrite( f"/home/juni/사진/flood/infer/{k}.jpg", seg_image)
292	308
293	309	# Show the image_binary
294	310	# cv2.imshow('Detections', img)
295	311	# cv2.waitKey(0)
296	312	# cv2.destroyAllWindows()
297	313
298		- # If you also want to display segmentation maps, you would need additional handling here
299		- # Example for displaying first mask if available:
300		- if len(mask_maps) != 0:
301	314
302		- seg_image = overlay_mask(img, mask_maps[:,:,0], color=(0, 255, 0), alpha=0.3)
303		- cv2.imshow("segmentation", seg_image)
304		- cv2.waitKey(0)
305		- cv2.destroyAllWindows()
306	315
307	316
308	317	if __name__ == "__main__":

Delete comment