-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathcanny2image_TRT.py
161 lines (123 loc) · 7.78 KB
/
canny2image_TRT.py
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
from share import *
import config
import cv2
import einops
import gradio as gr
import numpy as np
import torch
import random
from pytorch_lightning import seed_everything
from annotator.util import resize_image, HWC3
from annotator.canny import CannyDetector
from cldm.model import create_model, load_state_dict
from ddim_hacked_trt import DDIMSampler
from ddim_hacked_trt import *
from transformers import T5Tokenizer, T5EncoderModel, CLIPTokenizer, CLIPTextModel
from trt_util import *
import time
class hackathon():
def initialize(self):
self.apply_canny = CannyDetector()
self.model = create_model('./models/cldm_v15.yaml').cpu()
# self.model.load_state_dict(load_state_dict('/home/player/ControlNet/models/control_sd15_canny.pth', location='cuda'))
self.model = self.model.cuda()
self.device = torch.device("cuda")
self.tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14")
# # # self.controlnet_context = get_engine("./controlnet.plan")
# # # self.unet_context = get_engine("./unet.plan")
# self.combine_context = get_engine("./combine.plan")
# self.postnet_context = get_engine("./postnet.plan")
# self.ddim_sampler = DDIMSampler(self.model,controlnet_context=self.controlnet_context,unet_context=self.unet_context,postnet_context=self.postnet_context)
# clip_encoder
self.clip_engine = Engine("./clip_encoder.plan")
self.clip_engine.load()
self.clip_engine.activate(reuse_device_memory=None)
self.clip_engine.allocate_buffers()
# self.clip_context = get_engine("./clip_encoder.plan",dynamic=True)
# combine
self.combine_engine = Engine("./combine.plan")
self.combine_engine.load()
self.combine_engine.activate(reuse_device_memory=None)
self.combine_engine.allocate_buffers()
#vae
self.vae_engine = Engine("./vae_decoder.plan")
self.vae_engine.load()
self.vae_engine.activate(reuse_device_memory=None)
self.vae_engine.allocate_buffers()
# self.vae_context = get_engine("./vae_decoder.plan")
# #post
# self.post_engine = Engine("./postnet.plan")
# self.post_engine.load()
# self.post_engine.activate(reuse_device_memory=None)
# self.post_engine.allocate_buffers()
self.post_engine = get_engine("./postnet.plan")
self.ddim_sampler = DDIMSampler(self.model,combine_context=self.combine_engine,postnet_context=self.post_engine)
self.model.control_scales = [1]*13
def process(self, input_image, prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps, guess_mode, strength, scale, seed, eta, low_threshold, high_threshold):
ddim_steps = 10 # 为了测试想法
with torch.no_grad():
img = resize_image(HWC3(input_image), image_resolution)
H, W, C = img.shape
detected_map = self.apply_canny(img, low_threshold, high_threshold)
detected_map = HWC3(detected_map)
control = torch.from_numpy(detected_map.copy()).float().cuda() / 255.0
control = torch.stack([control for _ in range(num_samples)], dim=0)
control = einops.rearrange(control, 'b h w c -> b c h w').clone()
if seed == -1:
seed = random.randint(0, 65535)
seed_everything(seed)
# if config.save_memory:
# self.model.low_vram_shift(is_diffusing=False)
# # 调用clip encoder
# cond_text = [prompt + ', ' + a_prompt]
# cond_batch_encoding = self.tokenizer(cond_text, truncation=True, max_length=77, return_length=True,
# return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
# cond_tokens = cond_batch_encoding["input_ids"].to(self.device)
# uncond_text =[n_prompt]
# uncond_batch_encoding = self.tokenizer(uncond_text, truncation=True, max_length=77, return_length=True,
# return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
# uncond_tokens = uncond_batch_encoding["input_ids"].to(self.device)
# buffer_cilp_D = [cond_tokens.data_ptr(),self.ddim_sampler.bufferD["clip_encoder"]]
# self.clip_context.execute_v2(buffer_cilp_D)
# buffer_cilp_D1 = [uncond_tokens.data_ptr(),self.ddim_sampler.bufferD["clip_encoder1"]]
# self.clip_context.execute_v2(buffer_cilp_D1)
# cond = {"c_concat": [control], "c_crossattn": [self.ddim_sampler.bufferD["clip_encoder"]]}
# un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": [self.ddim_sampler.bufferD["clip_encoder1"]]}
# cond = {"c_concat": [control], "c_crossattn": [self.model.get_learned_conditioning([prompt + ', ' + a_prompt] * num_samples)]}
# un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": [self.model.get_learned_conditioning([n_prompt] * num_samples)]}
# -----clip的trt调用
cond_text = [prompt + ', ' + a_prompt, n_prompt]
cond_batch_encoding = self.tokenizer(cond_text, truncation=True, max_length=77, return_length=True,
return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
cond_tokens = cond_batch_encoding["input_ids"].type(torch.int32).to(self.device) #<---关键在这！
#-------------替换为异步
# buffer_cilp_D = [cond_tokens.data_ptr(),self.ddim_sampler.bufferD["clip_encoder"]]
# self.clip_context.execute_v2(buffer_cilp_D)
text_embeddings = runEngine(self.clip_engine, {"input_ids": cond_tokens},self.ddim_sampler.stream)['clip_encoder'].clone()
shape = (4, H // 8, W // 8)
# cond = self.model.get_learned_conditioning([prompt + ', ' + a_prompt, n_prompt])
control = torch.cat([control],1)
control = torch.cat([control,control],axis=0)
# cond = {"c_concat": [control], "c_crossattn": [cond]}
cond = {"c_concat": [control], "c_crossattn": [text_embeddings]}
un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": None}
# if config.save_memory:
# self.model.low_vram_shift(is_diffusing=True)
# self.model.control_scales = [strength * (0.825 ** float(12 - i)) for i in range(13)] if guess_mode else ([strength] * 13) # Magic number. IDK why. Perhaps because 0.825**12<0.01 but 0.826**12>0.01
samples, intermediates = self.ddim_sampler.sample(ddim_steps, num_samples,
shape, cond, verbose=False, eta=eta,
unconditional_guidance_scale=scale,
unconditional_conditioning=un_cond)
# # vae
# a = time.time()
# buffer_vae_D = [samples.data_ptr(),self.ddim_sampler.bufferD["dec"]]
# self.vae_context.execute_v2(buffer_vae_D)
# cudart.cudaMemcpy(self.ddim_sampler.bufferH["dec"].ctypes.data, self.ddim_sampler.bufferD["dec"],
# self.ddim_sampler.bufferH["dec"].nbytes, cudart.cudaMemcpyKind.cudaMemcpyDeviceToHost)
# results = [self.ddim_sampler.bufferH["dec"].transpose((0,2,3,1))[i] for i in range(num_samples)]
# b = time.time()
# print((b-a)*1000)
# vae cuda stream
buffer_vae = runEngine(self.vae_engine, {"image": samples},self.ddim_sampler.stream)['dec'].clone()
results = [einops.rearrange(buffer_vae, 'b c h w -> b h w c') .cpu().numpy()[0]]
return results