File size: 6,435 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
# 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 unittest

import numpy as np
import torch
from PIL import Image
from transformers import CLIPTokenizer
from transformers.models.blip_2.configuration_blip_2 import Blip2Config
from transformers.models.clip.configuration_clip import CLIPTextConfig

from diffusers import AutoencoderKL, BlipDiffusionPipeline, PNDMScheduler, UNet2DConditionModel
from diffusers.utils.testing_utils import enable_full_determinism
from src.diffusers.pipelines.blip_diffusion.blip_image_processing import BlipImageProcessor
from src.diffusers.pipelines.blip_diffusion.modeling_blip2 import Blip2QFormerModel
from src.diffusers.pipelines.blip_diffusion.modeling_ctx_clip import ContextCLIPTextModel

from ..test_pipelines_common import PipelineTesterMixin


enable_full_determinism()


class BlipDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
    pipeline_class = BlipDiffusionPipeline
    params = [
        "prompt",
        "reference_image",
        "source_subject_category",
        "target_subject_category",
    ]
    batch_params = [
        "prompt",
        "reference_image",
        "source_subject_category",
        "target_subject_category",
    ]
    required_optional_params = [
        "generator",
        "height",
        "width",
        "latents",
        "guidance_scale",
        "num_inference_steps",
        "neg_prompt",
        "guidance_scale",
        "prompt_strength",
        "prompt_reps",
    ]

    def get_dummy_components(self):
        torch.manual_seed(0)
        text_encoder_config = CLIPTextConfig(
            vocab_size=1000,
            hidden_size=8,
            intermediate_size=8,
            projection_dim=8,
            num_hidden_layers=1,
            num_attention_heads=1,
            max_position_embeddings=77,
        )
        text_encoder = ContextCLIPTextModel(text_encoder_config)

        vae = AutoencoderKL(
            in_channels=4,
            out_channels=4,
            down_block_types=("DownEncoderBlock2D",),
            up_block_types=("UpDecoderBlock2D",),
            block_out_channels=(8,),
            norm_num_groups=8,
            layers_per_block=1,
            act_fn="silu",
            latent_channels=4,
            sample_size=8,
        )

        blip_vision_config = {
            "hidden_size": 8,
            "intermediate_size": 8,
            "num_hidden_layers": 1,
            "num_attention_heads": 1,
            "image_size": 224,
            "patch_size": 14,
            "hidden_act": "quick_gelu",
        }

        blip_qformer_config = {
            "vocab_size": 1000,
            "hidden_size": 8,
            "num_hidden_layers": 1,
            "num_attention_heads": 1,
            "intermediate_size": 8,
            "max_position_embeddings": 512,
            "cross_attention_frequency": 1,
            "encoder_hidden_size": 8,
        }
        qformer_config = Blip2Config(
            vision_config=blip_vision_config,
            qformer_config=blip_qformer_config,
            num_query_tokens=8,
            tokenizer="hf-internal-testing/tiny-random-bert",
        )
        qformer = Blip2QFormerModel(qformer_config)

        unet = UNet2DConditionModel(
            block_out_channels=(8, 16),
            norm_num_groups=8,
            layers_per_block=1,
            sample_size=16,
            in_channels=4,
            out_channels=4,
            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
            cross_attention_dim=8,
        )
        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")

        scheduler = PNDMScheduler(
            beta_start=0.00085,
            beta_end=0.012,
            beta_schedule="scaled_linear",
            set_alpha_to_one=False,
            skip_prk_steps=True,
        )

        vae.eval()
        qformer.eval()
        text_encoder.eval()

        image_processor = BlipImageProcessor()

        components = {
            "text_encoder": text_encoder,
            "vae": vae,
            "qformer": qformer,
            "unet": unet,
            "tokenizer": tokenizer,
            "scheduler": scheduler,
            "image_processor": image_processor,
        }
        return components

    def get_dummy_inputs(self, device, seed=0):
        np.random.seed(seed)
        reference_image = np.random.rand(32, 32, 3) * 255
        reference_image = Image.fromarray(reference_image.astype("uint8")).convert("RGBA")

        if str(device).startswith("mps"):
            generator = torch.manual_seed(seed)
        else:
            generator = torch.Generator(device=device).manual_seed(seed)
        inputs = {
            "prompt": "swimming underwater",
            "generator": generator,
            "reference_image": reference_image,
            "source_subject_category": "dog",
            "target_subject_category": "dog",
            "height": 32,
            "width": 32,
            "guidance_scale": 7.5,
            "num_inference_steps": 2,
            "output_type": "np",
        }
        return inputs

    def test_blipdiffusion(self):
        device = "cpu"
        components = self.get_dummy_components()

        pipe = self.pipeline_class(**components)
        pipe = pipe.to(device)

        pipe.set_progress_bar_config(disable=None)

        image = pipe(**self.get_dummy_inputs(device))[0]
        image_slice = image[0, -3:, -3:, 0]

        assert image.shape == (1, 16, 16, 4)

        expected_slice = np.array(
            [0.5329548, 0.8372512, 0.33269387, 0.82096875, 0.43657133, 0.3783, 0.5953028, 0.51934963, 0.42142007]
        )

        assert (
            np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
        ), f" expected_slice {image_slice.flatten()}, but got {image_slice.flatten()}"