lunny/keras
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Adding back changes to view diff

Browse Source
This commit is contained in:
divyashreepathihalli 2023-09-26 14:53:11 +00:00 committed by François Chollet
parent bb70a8f0b2
commit c04da8144f
1 changed files with 30 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
examples/keras_io/vision/latent_walk.py
View File

@ -65,9 +65,9 @@ pip install keras-cv --upgrade --quiet
""" """
import keras_cv import keras_cv
from tensorflow import keras from keras_cv.backend import keras
import matplotlib.pyplot as plt import matplotlib.pyplot as plt
import tensorflow as tf from keras_cv.backend import ops
import numpy as np import numpy as np
import math import math
from PIL import Image from PIL import Image
@ -96,10 +96,10 @@ prompt_1 = "A watercolor painting of a Golden Retriever at the beach"
prompt_2 = "A still life DSLR photo of a bowl of fruit" prompt_2 = "A still life DSLR photo of a bowl of fruit"
interpolation_steps = 5 interpolation_steps = 5
encoding_1 = tf.squeeze(model.encode_text(prompt_1)) encoding_1 = ops.squeeze(model.encode_text(prompt_1))
encoding_2 = tf.squeeze(model.encode_text(prompt_2)) encoding_2 = ops.squeeze(model.encode_text(prompt_2))
interpolated_encodings = tf.linspace(encoding_1, encoding_2, interpolation_steps) interpolated_encodings = ops.linspace(encoding_1, encoding_2, interpolation_steps)
# Show the size of the latent manifold # Show the size of the latent manifold
print(f"Encoding shape: {encoding_1.shape}") print(f"Encoding shape: {encoding_1.shape}")
@ -111,7 +111,7 @@ keep the diffusion noise constant between images.
""" """
seed = 12345 seed = 12345
noise = tf.random.normal((512 // 8, 512 // 8, 4), seed=seed) noise = keras.random.normal((512 // 8, 512 // 8, 4), seed=seed)
images = model.generate_image( images = model.generate_image(
interpolated_encodings, interpolated_encodings,
@ -174,8 +174,8 @@ interpolation_steps = 150
batch_size = 3 batch_size = 3
batches = interpolation_steps // batch_size batches = interpolation_steps // batch_size
interpolated_encodings = tf.linspace(encoding_1, encoding_2, interpolation_steps) interpolated_encodings = ops.linspace(encoding_1, encoding_2, interpolation_steps)
batched_encodings = tf.split(interpolated_encodings, batches) batched_encodings = ops.split(interpolated_encodings, batches)
images = [] images = []
for batch in range(batches): for batch in range(batches):
@ -210,20 +210,20 @@ interpolation_steps = 6
batch_size = 3 batch_size = 3
batches = (interpolation_steps**2) // batch_size batches = (interpolation_steps**2) // batch_size
encoding_1 = tf.squeeze(model.encode_text(prompt_1)) encoding_1 = ops.squeeze(model.encode_text(prompt_1))
encoding_2 = tf.squeeze(model.encode_text(prompt_2)) encoding_2 = ops.squeeze(model.encode_text(prompt_2))
encoding_3 = tf.squeeze(model.encode_text(prompt_3)) encoding_3 = ops.squeeze(model.encode_text(prompt_3))
encoding_4 = tf.squeeze(model.encode_text(prompt_4)) encoding_4 = ops.squeeze(model.encode_text(prompt_4))
interpolated_encodings = tf.linspace( interpolated_encodings = ops.linspace(
tf.linspace(encoding_1, encoding_2, interpolation_steps), ops.linspace(encoding_1, encoding_2, interpolation_steps),
tf.linspace(encoding_3, encoding_4, interpolation_steps), ops.linspace(encoding_3, encoding_4, interpolation_steps),
interpolation_steps, interpolation_steps,
) )
interpolated_encodings = tf.reshape( interpolated_encodings = ops.reshape(
interpolated_encodings, (interpolation_steps**2, 77, 768) interpolated_encodings, (interpolation_steps**2, 77, 768)
) )
batched_encodings = tf.split(interpolated_encodings, batches) batched_encodings = ops.split(interpolated_encodings, batches)
images = [] images = []
for batch in range(batches): for batch in range(batches):
@ -293,18 +293,18 @@ batch_size = 3
batches = walk_steps // batch_size batches = walk_steps // batch_size
step_size = 0.005 step_size = 0.005
encoding = tf.squeeze( encoding = ops.squeeze(
model.encode_text("The Eiffel Tower in the style of starry night") model.encode_text("The Eiffel Tower in the style of starry night")
) )
# Note that (77, 768) is the shape of the text encoding. # Note that (77, 768) is the shape of the text encoding.
delta = tf.ones_like(encoding) * step_size delta = ops.ones_like(encoding) * step_size
walked_encodings = [] walked_encodings = []
for step_index in range(walk_steps): for step_index in range(walk_steps):
walked_encodings.append(encoding) walked_encodings.append(encoding)
encoding += delta encoding += delta
walked_encodings = tf.stack(walked_encodings) walked_encodings = ops.stack(walked_encodings)
batched_encodings = tf.split(walked_encodings, batches) batched_encodings = ops.split(walked_encodings, batches)
images = [] images = []
for batch in range(batches): for batch in range(batches):
@ -342,20 +342,20 @@ we began our walk, so we get a "loopable" result!
""" """
prompt = "An oil paintings of cows in a field next to a windmill in Holland" prompt = "An oil paintings of cows in a field next to a windmill in Holland"
encoding = tf.squeeze(model.encode_text(prompt)) encoding = ops.squeeze(model.encode_text(prompt))
walk_steps = 150 walk_steps = 150
batch_size = 3 batch_size = 3
batches = walk_steps // batch_size batches = walk_steps // batch_size
walk_noise_x = tf.random.normal(noise.shape, dtype=tf.float64) walk_noise_x = keras.random.normal(noise.shape, dtype="float64")
walk_noise_y = tf.random.normal(noise.shape, dtype=tf.float64) walk_noise_y = keras.random.normal(noise.shape, dtype="float64")
walk_scale_x = tf.cos(tf.linspace(0, 2, walk_steps) * math.pi) walk_scale_x = ops.cos(ops.linspace(0, 2, walk_steps) * math.pi)
walk_scale_y = tf.sin(tf.linspace(0, 2, walk_steps) * math.pi) walk_scale_y = ops.sin(ops.linspace(0, 2, walk_steps) * math.pi)
noise_x = tf.tensordot(walk_scale_x, walk_noise_x, axes=0) noise_x = ops.tensordot(walk_scale_x, walk_noise_x, axes=0)
noise_y = tf.tensordot(walk_scale_y, walk_noise_y, axes=0) noise_y = ops.tensordot(walk_scale_y, walk_noise_y, axes=0)
noise = tf.add(noise_x, noise_y) noise = ops.add(noise_x, noise_y)
batched_noise = tf.split(noise, batches) batched_noise = ops.split(noise, batches)
images = [] images = []
for batch in range(batches): for batch in range(batches):