Boa - 5 months ago 39

Python Question

I wrote a function which generates 2 coloured image blocks:

`def generate_block():`

x = np.ones((50, 50, 3))

x[:,:,0:3] = np.random.uniform(0, 1, (3,))

show_image(x)

y = np.ones((50, 50, 3))

y[:, :, 0:3] = np.random.uniform(0, 1, (3,))

show_image(y)

I would then like to combine those two colours to form a gradient, ie 1 image going from one colour to the other. I'm not sure how to continue, any advice? Using np.linspace() I can form a 1D array of steps but what then?

Answer

Is this what you are looking for ?

```
def generate_block():
x = np.ones((50, 50, 3))
x[:, :, 0:3] = np.random.uniform(0, 1, (3,))
plt.imshow(x)
plt.figure()
y = np.ones((50, 50, 3))
y[:,:,0:3] = np.random.uniform(0, 1, (3,))
plt.imshow(y)
plt.figure()
c = np.linspace(0, 1, 50)[:, None, None]
gradient = x + (y - x) * c
plt.imshow(gradient)
return x, y, gradient
```

To use `np.linspace`

as you suggested, I've used broadcasting which is a very powerful tool in numpy; read more here.

`c = np.linspace(0, 1, 50)`

creates an array of shape `(50,)`

with 50 numbers from 0 to 1, evenly spaced. Adding `[:, None, None]`

makes this array 3D, of shape `(50, 1, 1)`

. When using it in `(x - y) * c`

, since `x - y`

is `(50, 50, 3)`

, broadcasting happens for the last 2 dimensions. c is treated as an array we'll call d of shape `(50, 50, 3)`

, such that for i in range(50), `d[i, :, :]`

is an array of shape `(50, 3)`

filled with `c[i]`

.

so the first line of gradient is `x[0, :, :] + c[0] * (x[0, :, :] - y[0, :, :])`

, which is just `x[0, :, :]`

The second line is `x[1, :, :] + c[1] * (x[1, :, :] - y[1, :, :])`

, etc. The `i`

th line is the barycenter of `x[i]`

and `y[i]`

with coefficients `1 - c[i]`

and `c[i]`

You can do column-wise variation with [None, :, None] in the definition of c.