File size: 8,222 Bytes
8206b09 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 |
# coding=utf-8
# Copyright 2023 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import gc
import random
import unittest
import numpy as np
import torch
from diffusers import DDIMScheduler, KandinskyV22Pipeline, KandinskyV22PriorPipeline, UNet2DConditionModel, VQModel
from diffusers.utils import floats_tensor, load_numpy, slow, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu
from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
enable_full_determinism()
class Dummies:
@property
def text_embedder_hidden_size(self):
return 32
@property
def time_input_dim(self):
return 32
@property
def block_out_channels_0(self):
return self.time_input_dim
@property
def time_embed_dim(self):
return self.time_input_dim * 4
@property
def cross_attention_dim(self):
return 32
@property
def dummy_unet(self):
torch.manual_seed(0)
model_kwargs = {
"in_channels": 4,
# Out channels is double in channels because predicts mean and variance
"out_channels": 8,
"addition_embed_type": "image",
"down_block_types": ("ResnetDownsampleBlock2D", "SimpleCrossAttnDownBlock2D"),
"up_block_types": ("SimpleCrossAttnUpBlock2D", "ResnetUpsampleBlock2D"),
"mid_block_type": "UNetMidBlock2DSimpleCrossAttn",
"block_out_channels": (self.block_out_channels_0, self.block_out_channels_0 * 2),
"layers_per_block": 1,
"encoder_hid_dim": self.text_embedder_hidden_size,
"encoder_hid_dim_type": "image_proj",
"cross_attention_dim": self.cross_attention_dim,
"attention_head_dim": 4,
"resnet_time_scale_shift": "scale_shift",
"class_embed_type": None,
}
model = UNet2DConditionModel(**model_kwargs)
return model
@property
def dummy_movq_kwargs(self):
return {
"block_out_channels": [32, 64],
"down_block_types": ["DownEncoderBlock2D", "AttnDownEncoderBlock2D"],
"in_channels": 3,
"latent_channels": 4,
"layers_per_block": 1,
"norm_num_groups": 8,
"norm_type": "spatial",
"num_vq_embeddings": 12,
"out_channels": 3,
"up_block_types": [
"AttnUpDecoderBlock2D",
"UpDecoderBlock2D",
],
"vq_embed_dim": 4,
}
@property
def dummy_movq(self):
torch.manual_seed(0)
model = VQModel(**self.dummy_movq_kwargs)
return model
def get_dummy_components(self):
unet = self.dummy_unet
movq = self.dummy_movq
scheduler = DDIMScheduler(
num_train_timesteps=1000,
beta_schedule="linear",
beta_start=0.00085,
beta_end=0.012,
clip_sample=False,
set_alpha_to_one=False,
steps_offset=1,
prediction_type="epsilon",
thresholding=False,
)
components = {
"unet": unet,
"scheduler": scheduler,
"movq": movq,
}
return components
def get_dummy_inputs(self, device, seed=0):
image_embeds = floats_tensor((1, self.text_embedder_hidden_size), rng=random.Random(seed)).to(device)
negative_image_embeds = floats_tensor((1, self.text_embedder_hidden_size), rng=random.Random(seed + 1)).to(
device
)
if str(device).startswith("mps"):
generator = torch.manual_seed(seed)
else:
generator = torch.Generator(device=device).manual_seed(seed)
inputs = {
"image_embeds": image_embeds,
"negative_image_embeds": negative_image_embeds,
"generator": generator,
"height": 64,
"width": 64,
"guidance_scale": 4.0,
"num_inference_steps": 2,
"output_type": "np",
}
return inputs
class KandinskyV22PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
pipeline_class = KandinskyV22Pipeline
params = [
"image_embeds",
"negative_image_embeds",
]
batch_params = ["image_embeds", "negative_image_embeds"]
required_optional_params = [
"generator",
"height",
"width",
"latents",
"guidance_scale",
"num_inference_steps",
"return_dict",
"guidance_scale",
"num_images_per_prompt",
"output_type",
"return_dict",
]
test_xformers_attention = False
def get_dummy_inputs(self, device, seed=0):
dummies = Dummies()
return dummies.get_dummy_inputs(device=device, seed=seed)
def get_dummy_components(self):
dummies = Dummies()
return dummies.get_dummy_components()
def test_kandinsky(self):
device = "cpu"
components = self.get_dummy_components()
pipe = self.pipeline_class(**components)
pipe = pipe.to(device)
pipe.set_progress_bar_config(disable=None)
output = pipe(**self.get_dummy_inputs(device))
image = output.images
image_from_tuple = pipe(
**self.get_dummy_inputs(device),
return_dict=False,
)[0]
image_slice = image[0, -3:, -3:, -1]
image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.3420, 0.9505, 0.3919, 1.0000, 0.5188, 0.3109, 0.6139, 0.5624, 0.6811])
assert (
np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
), f" expected_slice {expected_slice}, but got {image_slice.flatten()}"
assert (
np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
), f" expected_slice {expected_slice}, but got {image_from_tuple_slice.flatten()}"
@slow
@require_torch_gpu
class KandinskyV22PipelineIntegrationTests(unittest.TestCase):
def tearDown(self):
# clean up the VRAM after each test
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def test_kandinsky_text2img(self):
expected_image = load_numpy(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
"/kandinskyv22/kandinskyv22_text2img_cat_fp16.npy"
)
pipe_prior = KandinskyV22PriorPipeline.from_pretrained(
"kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16
)
pipe_prior.to(torch_device)
pipeline = KandinskyV22Pipeline.from_pretrained(
"kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16
)
pipeline = pipeline.to(torch_device)
pipeline.set_progress_bar_config(disable=None)
prompt = "red cat, 4k photo"
generator = torch.Generator(device="cuda").manual_seed(0)
image_emb, zero_image_emb = pipe_prior(
prompt,
generator=generator,
num_inference_steps=5,
negative_prompt="",
).to_tuple()
generator = torch.Generator(device="cuda").manual_seed(0)
output = pipeline(
image_embeds=image_emb,
negative_image_embeds=zero_image_emb,
generator=generator,
num_inference_steps=100,
output_type="np",
)
image = output.images[0]
assert image.shape == (512, 512, 3)
assert_mean_pixel_difference(image, expected_image)
|