Commit @6e2a2791cb2ed2c3823c6bac85c198e6fb1101a6

윤영준 2023-07-12

data cleaning

@6e2a2791cb2ed2c3823c6bac85c198e6fb1101a6

f4b3691

6e2a279

.gitignore

--- .gitignore

+++ .gitignore

...	...	@@ -1,2 +1,5 @@
1	1	# this is image dataset that is about 9 GB
2		-data/source/Oxford_raindrop_dataset (파일 끝에 줄바꿈 문자 없음)
	2	+data/source/Oxford_raindrop_dataset
	3	+data/source/Peking_raindrop_dataset
	4	+data/source/Peking_raindrop_dataset/clean
	5	+data/source/Peking_raindrop_dataset/dirty (파일 끝에 줄바꿈 문자 없음)

f4b3691

6e2a279

binary mask map test.py

--- binary mask map test.py

+++ binary mask map test.py


 
     return shifted
 
-def binary_diff_mask(clean, dirty, thresold=0.3):
+def binary_diff_mask(clean, dirty, threshold=0.3):
     # this parts corrects gamma, and always remember, sRGB values are not in linear scale with lights intensity,
-    clean = np.power(clean, 1/2.2)
-    dirty = np.power(dirty, 1/2.2)
+    clean = np.power(clean, 2.2)
+    dirty = np.power(dirty, 2.2)
 
-    # averaged_per_pixel = np.abs(dirty / shift_img(clean, 5, 0) - 1)
+    averaged_per_pixel = np.abs(((dirty) / (clean+1)) - 1 )
     # print(averaged_per_pixel)
-    diff = np.abs(clean - dirty)
+    diff = (np.abs(dirty - clean)  * 0.8 + (dirty / clean -1)) * 0.2 / 2
 
-    bin_diff = (diff > thresold).astype(np.uint8)
+    # bin_diff = (diff > threshold).astype(np.uint8)
 
-    return bin_diff
+    return ((dirty / clean -1 ) > threshold ).astype(np.uint8)
 
 clean = glob.glob("data/source/Oxford_raindrop_dataset/clean/*.png")
 clean = sorted(clean)
 dirty = glob.glob("data/source/Oxford_raindrop_dataset/dirty/*.png")
 dirty = sorted(dirty)
 
-clean_img = cv2.imread(clean[34])
-dirty_img = cv2.imread(dirty[34])
+clean_img = cv2.imread(clean[0])
+dirty_img = cv2.imread(dirty[0])
 
-binary_diff_mask_img = binary_diff_mask(dirty_img/255, clean_img/255, thresold=0.05)
+# binary_diff_mask_img = binary_diff_mask(dirty_img/255, clean_img/255, threshold=0.05)
 
-k = 20
+k = 10
 
 def process(i, j):
     print(i)
     clean_img_copy = shift_img(clean_img, (i-k)/4, (j-k)/4)
-    binary_diff_mask_img = binary_diff_mask(dirty_img / 255, clean_img_copy / 255, threshold=0.3)
-    success = cv2.imwrite(f"test/test_img_x{(i-k)/4}-y{(j-k)/4}.png", binary_diff_mask_img*255)
-    if not success:
-        print(f"Failed to save image at test/test_img{(i - k) / 4}-{(j - k) / 4}.png")
+    binary_diff_mask_img = binary_diff_mask(dirty_img / 255, clean_img_copy / 255, threshold=0.2)
+    if not cv2.imwrite(f"test/test_img_x{(i-k)/4}-y{(j-k)/4}.png", binary_diff_mask_img*255):
+        raise Exception("image is not saved")
 
-
+# this thing does not throw error when error happens and just silently executes.... what?
 with concurrent.futures.ProcessPoolExecutor() as executor:
     for i in range(k*2):
         for j in range(k*2):
             executor.submit(process, i, j)
+cv2.imwrite(f"test/original.png", dirty_img)
+cv2.imwrite(f"test/original_c.png", clean_img)

6e2a279

data/source/Peking_raindrop_dataset/description.txt (added)

+++ data/source/Peking_raindrop_dataset/description.txt

...	...	@@ -0,0 +1,1 @@
	1	+https://github.com/rui1996/DeRaindrop

Add a comment

Open 0
Closed 0

List

...	...	@@ -12,41 +12,42 @@
12	12
13	13	return shifted
14	14
15		-def binary_diff_mask(clean, dirty, thresold=0.3):
	15	+def binary_diff_mask(clean, dirty, threshold=0.3):
16	16	# this parts corrects gamma, and always remember, sRGB values are not in linear scale with lights intensity,
17		- clean = np.power(clean, 1/2.2)
18		- dirty = np.power(dirty, 1/2.2)
	17	+ clean = np.power(clean, 2.2)
	18	+ dirty = np.power(dirty, 2.2)
19	19
20		- # averaged_per_pixel = np.abs(dirty / shift_img(clean, 5, 0) - 1)
	20	+ averaged_per_pixel = np.abs(((dirty) / (clean+1)) - 1 )
21	21	# print(averaged_per_pixel)
22		- diff = np.abs(clean - dirty)
	22	+ diff = (np.abs(dirty - clean) * 0.8 + (dirty / clean -1)) * 0.2 / 2
23	23
24		- bin_diff = (diff > thresold).astype(np.uint8)
	24	+ # bin_diff = (diff > threshold).astype(np.uint8)
25	25
26		- return bin_diff
	26	+ return ((dirty / clean -1 ) > threshold ).astype(np.uint8)
27	27
28	28	clean = glob.glob("data/source/Oxford_raindrop_dataset/clean/*.png")
29	29	clean = sorted(clean)
30	30	dirty = glob.glob("data/source/Oxford_raindrop_dataset/dirty/*.png")
31	31	dirty = sorted(dirty)
32	32
33		-clean_img = cv2.imread(clean[34])
34		-dirty_img = cv2.imread(dirty[34])
	33	+clean_img = cv2.imread(clean[0])
	34	+dirty_img = cv2.imread(dirty[0])
35	35
36		-binary_diff_mask_img = binary_diff_mask(dirty_img/255, clean_img/255, thresold=0.05)
	36	+# binary_diff_mask_img = binary_diff_mask(dirty_img/255, clean_img/255, threshold=0.05)
37	37
38		-k = 20
	38	+k = 10
39	39
40	40	def process(i, j):
41	41	print(i)
42	42	clean_img_copy = shift_img(clean_img, (i-k)/4, (j-k)/4)
43		- binary_diff_mask_img = binary_diff_mask(dirty_img / 255, clean_img_copy / 255, threshold=0.3)
44		- success = cv2.imwrite(f"test/test_img_x{(i-k)/4}-y{(j-k)/4}.png", binary_diff_mask_img*255)
45		- if not success:
46		- print(f"Failed to save image at test/test_img{(i - k) / 4}-{(j - k) / 4}.png")
	43	+ binary_diff_mask_img = binary_diff_mask(dirty_img / 255, clean_img_copy / 255, threshold=0.2)
	44	+ if not cv2.imwrite(f"test/test_img_x{(i-k)/4}-y{(j-k)/4}.png", binary_diff_mask_img*255):
	45	+ raise Exception("image is not saved")
47	46
48		-
	47	+# this thing does not throw error when error happens and just silently executes.... what?
49	48	with concurrent.futures.ProcessPoolExecutor() as executor:
50	49	for i in range(k*2):
51	50	for j in range(k*2):
52	51	executor.submit(process, i, j)
	52	+cv2.imwrite(f"test/original.png", dirty_img)
	53	+cv2.imwrite(f"test/original_c.png", clean_img) (파일 끝에 줄바꿈 문자 없음)

Delete comment