File size: 10,347 Bytes
43b7e92
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
# coding=utf-8
# Copyright 2024 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 PIL import Image
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer

from diffusers import AutoencoderKL, PNDMScheduler, StableDiffusionInpaintPipeline, UNet2DConditionModel
from diffusers.utils.testing_utils import (
    enable_full_determinism,
    floats_tensor,
    load_image,
    load_numpy,
    require_torch_gpu,
    slow,
    torch_device,
)

from ..pipeline_params import (
    TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS,
    TEXT_GUIDED_IMAGE_INPAINTING_PARAMS,
    TEXT_TO_IMAGE_CALLBACK_CFG_PARAMS,
)
from ..test_pipelines_common import PipelineKarrasSchedulerTesterMixin, PipelineLatentTesterMixin, PipelineTesterMixin


enable_full_determinism()


class StableDiffusion2InpaintPipelineFastTests(
    PipelineLatentTesterMixin, PipelineKarrasSchedulerTesterMixin, PipelineTesterMixin, unittest.TestCase
):
    pipeline_class = StableDiffusionInpaintPipeline
    params = TEXT_GUIDED_IMAGE_INPAINTING_PARAMS
    batch_params = TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS
    image_params = frozenset(
        []
    )  # TO-DO: update image_params once pipeline is refactored with VaeImageProcessor.preprocess
    image_latents_params = frozenset([])
    callback_cfg_params = TEXT_TO_IMAGE_CALLBACK_CFG_PARAMS.union({"mask", "masked_image_latents"})

    def get_dummy_components(self):
        torch.manual_seed(0)
        unet = UNet2DConditionModel(
            block_out_channels=(32, 64),
            layers_per_block=2,
            sample_size=32,
            in_channels=9,
            out_channels=4,
            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
            cross_attention_dim=32,
            # SD2-specific config below
            attention_head_dim=(2, 4),
            use_linear_projection=True,
        )
        scheduler = PNDMScheduler(skip_prk_steps=True)
        torch.manual_seed(0)
        vae = AutoencoderKL(
            block_out_channels=[32, 64],
            in_channels=3,
            out_channels=3,
            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
            latent_channels=4,
            sample_size=128,
        )
        torch.manual_seed(0)
        text_encoder_config = CLIPTextConfig(
            bos_token_id=0,
            eos_token_id=2,
            hidden_size=32,
            intermediate_size=37,
            layer_norm_eps=1e-05,
            num_attention_heads=4,
            num_hidden_layers=5,
            pad_token_id=1,
            vocab_size=1000,
            # SD2-specific config below
            hidden_act="gelu",
            projection_dim=512,
        )
        text_encoder = CLIPTextModel(text_encoder_config)
        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")

        components = {
            "unet": unet,
            "scheduler": scheduler,
            "vae": vae,
            "text_encoder": text_encoder,
            "tokenizer": tokenizer,
            "safety_checker": None,
            "feature_extractor": None,
            "image_encoder": None,
        }
        return components

    def get_dummy_inputs(self, device, seed=0):
        # TODO: use tensor inputs instead of PIL, this is here just to leave the old expected_slices untouched
        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
        image = image.cpu().permute(0, 2, 3, 1)[0]
        init_image = Image.fromarray(np.uint8(image)).convert("RGB").resize((64, 64))
        mask_image = Image.fromarray(np.uint8(image + 4)).convert("RGB").resize((64, 64))
        if str(device).startswith("mps"):
            generator = torch.manual_seed(seed)
        else:
            generator = torch.Generator(device=device).manual_seed(seed)
        inputs = {
            "prompt": "A painting of a squirrel eating a burger",
            "image": init_image,
            "mask_image": mask_image,
            "generator": generator,
            "num_inference_steps": 2,
            "guidance_scale": 6.0,
            "output_type": "np",
        }
        return inputs

    def test_stable_diffusion_inpaint(self):
        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
        components = self.get_dummy_components()
        sd_pipe = StableDiffusionInpaintPipeline(**components)
        sd_pipe = sd_pipe.to(device)
        sd_pipe.set_progress_bar_config(disable=None)

        inputs = self.get_dummy_inputs(device)
        image = sd_pipe(**inputs).images
        image_slice = image[0, -3:, -3:, -1]

        assert image.shape == (1, 64, 64, 3)
        expected_slice = np.array([0.4727, 0.5735, 0.3941, 0.5446, 0.5926, 0.4394, 0.5062, 0.4654, 0.4476])

        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

    def test_inference_batch_single_identical(self):
        super().test_inference_batch_single_identical(expected_max_diff=3e-3)


@slow
@require_torch_gpu
class StableDiffusionInpaintPipelineIntegrationTests(unittest.TestCase):
    def setUp(self):
        # clean up the VRAM before each test
        super().setUp()
        gc.collect()
        torch.cuda.empty_cache()

    def tearDown(self):
        # clean up the VRAM after each test
        super().tearDown()
        gc.collect()
        torch.cuda.empty_cache()

    def test_stable_diffusion_inpaint_pipeline(self):
        init_image = load_image(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
            "/sd2-inpaint/init_image.png"
        )
        mask_image = load_image(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png"
        )
        expected_image = load_numpy(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint"
            "/yellow_cat_sitting_on_a_park_bench.npy"
        )

        model_id = "stabilityai/stable-diffusion-2-inpainting"
        pipe = StableDiffusionInpaintPipeline.from_pretrained(model_id, safety_checker=None)
        pipe.to(torch_device)
        pipe.set_progress_bar_config(disable=None)
        pipe.enable_attention_slicing()

        prompt = "Face of a yellow cat, high resolution, sitting on a park bench"

        generator = torch.manual_seed(0)
        output = pipe(
            prompt=prompt,
            image=init_image,
            mask_image=mask_image,
            generator=generator,
            output_type="np",
        )
        image = output.images[0]

        assert image.shape == (512, 512, 3)
        assert np.abs(expected_image - image).max() < 9e-3

    def test_stable_diffusion_inpaint_pipeline_fp16(self):
        init_image = load_image(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
            "/sd2-inpaint/init_image.png"
        )
        mask_image = load_image(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png"
        )
        expected_image = load_numpy(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint"
            "/yellow_cat_sitting_on_a_park_bench_fp16.npy"
        )

        model_id = "stabilityai/stable-diffusion-2-inpainting"
        pipe = StableDiffusionInpaintPipeline.from_pretrained(
            model_id,
            torch_dtype=torch.float16,
            safety_checker=None,
        )
        pipe.to(torch_device)
        pipe.set_progress_bar_config(disable=None)
        pipe.enable_attention_slicing()

        prompt = "Face of a yellow cat, high resolution, sitting on a park bench"

        generator = torch.manual_seed(0)
        output = pipe(
            prompt=prompt,
            image=init_image,
            mask_image=mask_image,
            generator=generator,
            output_type="np",
        )
        image = output.images[0]

        assert image.shape == (512, 512, 3)
        assert np.abs(expected_image - image).max() < 5e-1

    def test_stable_diffusion_pipeline_with_sequential_cpu_offloading(self):
        torch.cuda.empty_cache()
        torch.cuda.reset_max_memory_allocated()
        torch.cuda.reset_peak_memory_stats()

        init_image = load_image(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
            "/sd2-inpaint/init_image.png"
        )
        mask_image = load_image(
            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd2-inpaint/mask.png"
        )

        model_id = "stabilityai/stable-diffusion-2-inpainting"
        pndm = PNDMScheduler.from_pretrained(model_id, subfolder="scheduler")
        pipe = StableDiffusionInpaintPipeline.from_pretrained(
            model_id,
            safety_checker=None,
            scheduler=pndm,
            torch_dtype=torch.float16,
        )
        pipe.set_progress_bar_config(disable=None)
        pipe.enable_attention_slicing(1)
        pipe.enable_sequential_cpu_offload()

        prompt = "Face of a yellow cat, high resolution, sitting on a park bench"

        generator = torch.manual_seed(0)
        _ = pipe(
            prompt=prompt,
            image=init_image,
            mask_image=mask_image,
            generator=generator,
            num_inference_steps=2,
            output_type="np",
        )

        mem_bytes = torch.cuda.max_memory_allocated()
        # make sure that less than 2.65 GB is allocated
        assert mem_bytes < 2.65 * 10**9