ValueError：无法将输入数组从形状（50,50,3）广播到形状（50,50）

Question

Trying to load data from dataset into an array data[ ], giving me a broadcasting error.

code:

path = os.path.join(cur_path , r'Pothole_Image_Data')
print(path)
images = os.listdir(path)
for i in images   :
  try   :
     image =  Image.open(path + '\\'  + i )
     image = image.resize((50,50))
     image = np.array(image)
     data.append(image)
  except : 
    print("Image not found")

data = np.array(data)
print(data.shape)

error:

  ValueError  Traceback (most recent call last)

   11         print("Image not found")
   ---> 13 data = np.array(data)
   14 print(data.shape)

ValueError: could not broadcast input array from shape (50,50,3) into shape (50,50)

Answer 1

Images are usually loaded in as X * Y * 3 (or 3 * X * Y) arrays, as images store values for the red, green and blue pixelvalues. So either reshape them into a 50,50,3 array (creating a 4D N * 50 * 50 *3) array, or think on how you want to merge the 3 color channels.

If you want to make a greyscale picture out of it (which can be stored in a 50x50 array), use for example luminosity method: val = (0.3 * R) + (0.59 * G) + (0.11 * B).

But this depends on your application. I assume that (given the way you construct your image array) that it's gonna be used in a Neural Network? Then I'd just use the 3 channels. Tensorflow can deal with that, Conv2d even has a variable to set the number of channels.