williamberman commited on
Commit
1c52cfa
1 Parent(s): a77d9c8

update README

Browse files
README.md CHANGED
@@ -18,301 +18,21 @@ Controlnet's auxiliary models are trained with stable diffusion 1.5. Experimenta
18
  The auxiliary conditioning is passed directly to the diffusers pipeline. If you want to process an image to create the auxiliary conditioning, external dependencies are required.
19
 
20
  Some of the additional conditionings can be extracted from images via additional models. We extracted these
21
- additional models from the original controlnet repo into a separate package that can be found on [github](https://github.com/patrickvonplaten/human_pose.git).
22
-
23
- ## Canny edge detection
24
-
25
- Install opencv
26
-
27
- ```sh
28
- $ pip install opencv-contrib-python
29
- ```
30
-
31
- ```python
32
- import cv2
33
- from PIL import Image
34
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
35
- import torch
36
- import numpy as np
37
-
38
- image = Image.open('images/bird.png')
39
- image = np.array(image)
40
-
41
- low_threshold = 100
42
- high_threshold = 200
43
-
44
- image = cv2.Canny(image, low_threshold, high_threshold)
45
- image = image[:, :, None]
46
- image = np.concatenate([image, image, image], axis=2)
47
- image = Image.fromarray(image)
48
-
49
- controlnet = ControlNetModel.from_pretrained(
50
- "fusing/stable-diffusion-v1-5-controlnet-canny",
51
- )
52
-
53
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
54
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
55
- )
56
- pipe.to('cuda')
57
-
58
- image = pipe("bird", image).images[0]
59
-
60
- image.save('images/bird_canny_out.png')
61
- ```
62
-
63
- ![bird](./images/bird.png)
64
-
65
- ![bird_canny](./images/bird_canny.png)
66
-
67
- ![bird_canny_out](./images/bird_canny_out.png)
68
-
69
- ## M-LSD Straight line detection
70
-
71
- Install the additional controlnet models package.
72
-
73
- ```sh
74
- $ pip install git+https://github.com/patrickvonplaten/human_pose.git
75
- ```
76
-
77
- ```py
78
- from PIL import Image
79
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
80
- import torch
81
- from human_pose import MLSDdetector
82
-
83
- mlsd = MLSDdetector.from_pretrained('lllyasviel/ControlNet')
84
-
85
- image = Image.open('images/room.png')
86
-
87
- image = mlsd(image)
88
-
89
- controlnet = ControlNetModel.from_pretrained(
90
- "fusing/stable-diffusion-v1-5-controlnet-mlsd",
91
- )
92
-
93
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
94
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
95
- )
96
- pipe.to('cuda')
97
-
98
- image = pipe("room", image).images[0]
99
-
100
- image.save('images/room_mlsd_out.png')
101
- ```
102
-
103
- ![room](./images/room.png)
104
-
105
- ![room_mlsd](./images/room_mlsd.png)
106
-
107
- ![room_mlsd_out](./images/room_mlsd_out.png)
108
-
109
- ## Pose estimation
110
-
111
- Install the additional controlnet models package.
112
-
113
- ```sh
114
- $ pip install git+https://github.com/patrickvonplaten/human_pose.git
115
- ```
116
-
117
- ```py
118
- from PIL import Image
119
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
120
- import torch
121
- from human_pose import OpenposeDetector
122
-
123
- openpose = OpenposeDetector.from_pretrained('lllyasviel/ControlNet')
124
-
125
- image = Image.open('images/pose.png')
126
-
127
- image = openpose(image)
128
-
129
- controlnet = ControlNetModel.from_pretrained(
130
- "fusing/stable-diffusion-v1-5-controlnet-openpose",
131
- )
132
-
133
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
134
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
135
- )
136
- pipe.to('cuda')
137
-
138
- image = pipe("chef in the kitchen", image).images[0]
139
-
140
- image.save('images/chef_pose_out.png')
141
- ```
142
-
143
- ![pose](./images/pose.png)
144
-
145
- ![openpose](./images/openpose.png)
146
-
147
- ![chef_pose_out](./images/chef_pose_out.png)
148
-
149
- ## Semantic Segmentation
150
-
151
- Semantic segmentation relies on transformers. Transformers is a
152
- dependency of diffusers for running controlnet, so you should
153
- have it installed already.
154
-
155
- ```py
156
- from transformers import AutoImageProcessor, UperNetForSemanticSegmentation
157
- from PIL import Image
158
- import numpy as np
159
- from controlnet_utils import ade_palette
160
- import torch
161
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
162
-
163
- image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-small")
164
- image_segmentor = UperNetForSemanticSegmentation.from_pretrained("openmmlab/upernet-convnext-small")
165
-
166
- image = Image.open("./images/house.png").convert('RGB')
167
-
168
- pixel_values = image_processor(image, return_tensors="pt").pixel_values
169
-
170
- with torch.no_grad():
171
- outputs = image_segmentor(pixel_values)
172
-
173
- seg = image_processor.post_process_semantic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
174
-
175
- color_seg = np.zeros((seg.shape[0], seg.shape[1], 3), dtype=np.uint8) # height, width, 3
176
-
177
- palette = np.array(ade_palette())
178
-
179
- for label, color in enumerate(palette):
180
- color_seg[seg == label, :] = color
181
-
182
- color_seg = color_seg.astype(np.uint8)
183
-
184
- image = Image.fromarray(color_seg)
185
-
186
- controlnet = ControlNetModel.from_pretrained(
187
- "fusing/stable-diffusion-v1-5-controlnet-seg",
188
- )
189
-
190
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
191
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
192
- )
193
- pipe.to('cuda')
194
-
195
- image = pipe("house", image).images[0]
196
-
197
- image.save('./images/house_seg_out.png')
198
- ```
199
-
200
- ![house](images/house.png)
201
-
202
- ![house_seg](images/house_seg.png)
203
-
204
- ![house_seg_out](images/house_seg_out.png)
205
-
206
- ## Depth control
207
-
208
- Depth control relies on transformers. Transformers is a dependency of diffusers for running controlnet, so
209
- you should have it installed already.
210
-
211
- ```py
212
- from transformers import pipeline
213
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
214
- from PIL import Image
215
- import numpy as np
216
-
217
- depth_estimator = pipeline('depth-estimation')
218
-
219
- image = Image.open('./images/stormtrooper.png')
220
- image = depth_estimator(image)['depth']
221
- image = np.array(image)
222
- image = image[:, :, None]
223
- image = np.concatenate([image, image, image], axis=2)
224
- image = Image.fromarray(image)
225
-
226
- controlnet = ControlNetModel.from_pretrained(
227
- "fusing/stable-diffusion-v1-5-controlnet-depth",
228
- )
229
-
230
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
231
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
232
- )
233
- pipe.to('cuda')
234
-
235
- image = pipe("Stormtrooper's lecture", image).images[0]
236
-
237
- image.save('./images/stormtrooper_depth_out.png')
238
- ```
239
-
240
- ![stormtrooper](./images/stormtrooper.png)
241
-
242
- ![stormtrooler_depth](./images/stormtrooper_depth.png)
243
-
244
- ![stormtrooler_depth_out](./images/stormtrooper_depth_out.png)
245
-
246
-
247
- ## Normal map
248
-
249
- ```py
250
- from PIL import Image
251
- from transformers import pipeline
252
- import numpy as np
253
- import cv2
254
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
255
-
256
- image = Image.open("images/toy.png").convert("RGB")
257
-
258
- depth_estimator = pipeline("depth-estimation", model ="Intel/dpt-hybrid-midas" )
259
-
260
- image = depth_estimator(image)['predicted_depth'][0]
261
-
262
- image = image.numpy()
263
-
264
- image_depth = image.copy()
265
- image_depth -= np.min(image_depth)
266
- image_depth /= np.max(image_depth)
267
-
268
- bg_threhold = 0.4
269
-
270
- x = cv2.Sobel(image, cv2.CV_32F, 1, 0, ksize=3)
271
- x[image_depth < bg_threhold] = 0
272
-
273
- y = cv2.Sobel(image, cv2.CV_32F, 0, 1, ksize=3)
274
- y[image_depth < bg_threhold] = 0
275
-
276
- z = np.ones_like(x) * np.pi * 2.0
277
-
278
- image = np.stack([x, y, z], axis=2)
279
- image /= np.sum(image ** 2.0, axis=2, keepdims=True) ** 0.5
280
- image = (image * 127.5 + 127.5).clip(0, 255).astype(np.uint8)
281
- image = Image.fromarray(image)
282
-
283
- controlnet = ControlNetModel.from_pretrained(
284
- "fusing/stable-diffusion-v1-5-controlnet-normal",
285
- )
286
-
287
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
288
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
289
- )
290
- pipe.to('cuda')
291
-
292
- image = pipe("cute toy", image).images[0]
293
-
294
- image.save('images/toy_normal_out.png')
295
- ```
296
-
297
- ![toy](./images/toy.png)
298
-
299
- ![toy_normal](./images/toy_normal.png)
300
-
301
- ![toy_normal_out](./images/toy_normal_out.png)
302
 
303
  ## Scribble
304
 
305
  Install the additional controlnet models package.
306
 
307
  ```sh
308
- $ pip install git+https://github.com/patrickvonplaten/human_pose.git
309
  ```
310
 
311
  ```py
312
  from PIL import Image
313
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
314
  import torch
315
- from human_pose import HEDdetector
316
 
317
  hed = HEDdetector.from_pretrained('lllyasviel/ControlNet')
318
 
@@ -321,15 +41,23 @@ image = Image.open('images/bag.png')
321
  image = hed(image, scribble=True)
322
 
323
  controlnet = ControlNetModel.from_pretrained(
324
- "fusing/stable-diffusion-v1-5-controlnet-scribble",
325
  )
326
 
327
  pipe = StableDiffusionControlNetPipeline.from_pretrained(
328
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
329
  )
330
- pipe.to('cuda')
331
 
332
- image = pipe("bag", image).images[0]
 
 
 
 
 
 
 
 
 
333
 
334
  image.save('images/bag_scribble_out.png')
335
  ```
@@ -340,42 +68,6 @@ image.save('images/bag_scribble_out.png')
340
 
341
  ![bag_scribble_out](./images/bag_scribble_out.png)
342
 
343
- ## HED Boundary
344
-
345
- Install the additional controlnet models package.
346
-
347
- ```sh
348
- $ pip install git+https://github.com/patrickvonplaten/human_pose.git
349
- ```
350
-
351
- ```py
352
- from PIL import Image
353
- from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
354
- import torch
355
- from human_pose import HEDdetector
356
-
357
- hed = HEDdetector.from_pretrained('lllyasviel/ControlNet')
358
-
359
- image = Image.open('images/man.png')
360
-
361
- image = hed(image)
362
-
363
- controlnet = ControlNetModel.from_pretrained(
364
- "fusing/stable-diffusion-v1-5-controlnet-hed",
365
- )
366
-
367
- pipe = StableDiffusionControlNetPipeline.from_pretrained(
368
- "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None
369
- )
370
- pipe.to('cuda')
371
-
372
- image = pipe("oil painting of handsome old man, masterpiece", image).images[0]
373
-
374
- image.save('images/man_hed_out.png')
375
- ```
376
-
377
- ![man](./images/man.png)
378
-
379
- ![man_hed](./images/man_hed.png)
380
 
381
- ![man_hed_out](./images/man_hed_out.png)
18
  The auxiliary conditioning is passed directly to the diffusers pipeline. If you want to process an image to create the auxiliary conditioning, external dependencies are required.
19
 
20
  Some of the additional conditionings can be extracted from images via additional models. We extracted these
21
+ additional models from the original controlnet repo into a separate package that can be found on [github](https://github.com/patrickvonplaten/controlnet_aux.git).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
 
23
  ## Scribble
24
 
25
  Install the additional controlnet models package.
26
 
27
  ```sh
28
+ $ pip install git+https://github.com/patrickvonplaten/controlnet_aux.git
29
  ```
30
 
31
  ```py
32
  from PIL import Image
33
+ from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
34
  import torch
35
+ from controlnet_aux import HEDdetector
36
 
37
  hed = HEDdetector.from_pretrained('lllyasviel/ControlNet')
38
 
41
  image = hed(image, scribble=True)
42
 
43
  controlnet = ControlNetModel.from_pretrained(
44
+ "fusing/stable-diffusion-v1-5-controlnet-scribble", torch_dtype=torch.float16
45
  )
46
 
47
  pipe = StableDiffusionControlNetPipeline.from_pretrained(
48
+ "runwayml/stable-diffusion-v1-5", controlnet=controlnet, safety_checker=None, torch_dtype=torch.float16
49
  )
 
50
 
51
+ pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
52
+
53
+ # Remove if you do not have xformers installed
54
+ # see https://huggingface.co/docs/diffusers/v0.13.0/en/optimization/xformers#installing-xformers
55
+ # for installation instructions
56
+ pipe.enable_xformers_memory_efficient_attention()
57
+
58
+ pipe.enable_model_cpu_offload()
59
+
60
+ image = pipe("bag", image, num_inference_steps=20).images[0]
61
 
62
  image.save('images/bag_scribble_out.png')
63
  ```
68
 
69
  ![bag_scribble_out](./images/bag_scribble_out.png)
70
 
71
+ ### Training
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
72
 
73
+ The scribble model was trained on 500k scribble-image, caption pairs. The scribble images were generated with HED boundary detection and a set of data augmentations — thresholds, masking, morphological transformations, and non-maximum suppression. The model was trained for 150 GPU-hours with Nvidia A100 80G using the canny model as a base model.
controlnet_utils.py DELETED
@@ -1,40 +0,0 @@
1
- def ade_palette():
2
- """ADE20K palette that maps each class to RGB values."""
3
- return [[120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50],
4
- [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255],
5
- [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7],
6
- [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82],
7
- [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3],
8
- [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255],
9
- [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220],
10
- [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224],
11
- [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255],
12
- [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7],
13
- [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153],
14
- [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255],
15
- [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0],
16
- [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255],
17
- [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255],
18
- [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255],
19
- [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0],
20
- [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0],
21
- [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255],
22
- [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255],
23
- [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20],
24
- [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255],
25
- [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255],
26
- [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255],
27
- [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0],
28
- [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0],
29
- [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255],
30
- [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112],
31
- [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160],
32
- [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163],
33
- [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0],
34
- [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0],
35
- [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255],
36
- [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204],
37
- [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255],
38
- [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255],
39
- [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194],
40
- [102, 255, 0], [92, 0, 255]]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
images/bag_scribble_out.png CHANGED
images/bird.png DELETED

Git LFS Details

  • SHA256: cad49fc7d3071b2bcd078bc8dde365f8fa62eaa6d43705fd50c212794a3aac35
  • Pointer size: 132 Bytes
  • Size of remote file: 1.07 MB
images/bird_canny.png DELETED
Binary file (29.1 kB)
images/bird_canny_out.png DELETED
Binary file (845 kB)
images/chef_pose_out.png DELETED
Binary file (570 kB)
images/house.png DELETED
Binary file (391 kB)
images/house_seg.png DELETED
Binary file (3.68 kB)
images/house_seg_out.png DELETED
Binary file (472 kB)
images/man.png DELETED
Binary file (773 kB)
images/man_hed.png DELETED
Binary file (118 kB)
images/man_hed_out.png DELETED
Binary file (737 kB)
images/openpose.png DELETED
Binary file (6.55 kB)
images/pose.png DELETED
Binary file (592 kB)
images/room.png DELETED
Binary file (637 kB)
images/room_mlsd.png DELETED
Binary file (9.06 kB)
images/room_mlsd_out.png DELETED
Binary file (575 kB)
images/stormtrooper.png DELETED
Binary file (218 kB)
images/stormtrooper_depth.png DELETED
Binary file (54.1 kB)
images/stormtrooper_depth_out.png DELETED
Binary file (343 kB)
images/toy.png DELETED
Binary file (312 kB)
images/toy_normal.png DELETED
Binary file (90.1 kB)
images/toy_normal_out.png DELETED
Binary file (231 kB)