Question

我正在尝试使用numpy将一组提供的坐标划分为Python 3中的几个存储桶。我有一个桶网格。见下文：

def partition(image, num_tiles):
    """Divide an image into a (num_tiles x num_tiles) grid and return the 
    partitioned input."""

    # The object to return. Ignore - I am just trying to test 'draw' works currently.
    partitioned_image = np.empty((num_tiles, num_tiles), dtype=object)

    draw = []

    # The input array contains coordinates of the form [xMin, xMax, yMin, yMax]. 
    # This is because these are coordinates for bounding boxes around biological cells. 
    # When I say 'point(s)', I refer to a [xMn, xMx, yMn, yMx] array(s).
    xMin = image[:,0]
    xMax = image[:,1]
    yMin = image[:,2]
    yMax = image[:,3]

    # The base to start searching from (not 0,0). 
    x_base = min(xMin)
    y_base = min(yMin)

    # max(?Max) - min(?Min) defines the entire range for the variable. Divide this 
    # range by the number of tiles, which is the number of ticks of the grid. 
    # E.g. range is 100, want a 10x10 grid, so we step along in steps of 10.
    x_step = (max(xMax) - min(xMin)) // num_tiles
    y_step = (max(yMax) - min(yMin)) // num_tiles


    for i in range(num_tiles):
        for j in range(num_tiles):
            # Define the bottom-left point of the region of interest (a tile)
            x_left = x_base + x_step * i
            y_low = y_base + y_step * j
            # Define the upper-right point of the region of interest
            x_right = x_base + x_step * (i + 1)
            y_high = y_base + y_step * (j + 1)

            # Every point in image that is within the region gets added to the
            # draw list. Remember, each point is of the form [xMn, xMx, yMn, yMx]
            result = ((yMin >= y_low) & (yMax < y_high) &
                  (xMin >= x_left) & (xMax < x_right)).nonzero()[0]

            for coordinates in image[result]:
                draw.append(coordinates)

            # I would want to add the actual points to my partitioned_input array
            # here, in the corresponding tile. The above code for draw is *JUST TESTING*.

    # Convert draw list to numpy array and check to see we got all the points.
    draw = np.asarray(draw)
    print(draw.shape == image.shape)  # We do not. This is annoying.

    # Below is the code for plotting. I just take the average of 
    # the xMin/yMin and xMax/yMax values for this.

    draw_xAvg = np.mean(np.array([draw[:,0], draw[:,1]]), axis=0)
    draw_yAvg = np.mean(np.array([draw[:,2], draw[:,3]]), axis=0)

    image_xAvg = np.mean(np.array([image[:,0], image[:,1]]), axis=0)
    image_yAvg = np.mean(np.array([image[:,2], image[:,3]]), axis=0)    

    fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize=(30, 10))

    ax1.set_title('Test', fontsize=30)
    ax1.scatter(draw_xAvg, draw_yAvg, s=0.1, c='b')

    ax2.set_title('Image', fontsize=30)
    ax2.scatter(image_xAvg, image_yAvg, s=0.1, c='r')

    ax3.set_title('Overlay (Image)', fontsize=30)
    ax3.scatter(image_xAvg, image_yAvg, s=0.1, c='r')
    ax3.scatter(draw_xAvg, draw_yAvg, s=0.1, c='b')

    # Would return this once I partitioned the input correctly.
    # The idea is to have a list per tile of all the points found in that tile.
    # All I am doing is checking that I get the right number of points in total.
    return partitioned_image

致电代码：

partitioned_cells = partition(cells, 20)

正如您所看到的，我按照与输入大小成比例的步骤逐步输入。这应该是完全正确的，我确实得到了绝大多数的要点，所以代码并没有完全错误，我的逻辑很好。但是，我希望在下面的第三个图中完美重叠：

如果你仔细观察右手图，你可以看到一个明显的网格状红色，蓝色不重叠，特别是在该图的右侧 - 所得到的numpy阵列的大小（12948 v 13804）也确认存在不匹配，红色数量超过蓝色。我在分区中遗漏了一些坐标。

我不知道为什么会这样 - 即使我的边界是包容性的（>=或<=），他们仍然无法得到所有的观点。我不明白为什么。有人可以解释或猜测为什么会这样吗？

将图像分割为网格时缺少坐标

0 个答案: