File size: 8,217 Bytes
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f2e0f04
3f6a58a
 
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
ec39fe8
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3f6a58a
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3f6a58a
a23872f
 
 
eac223c
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eac223c
a23872f
eac223c
 
a23872f
eac223c
a23872f
eac223c
 
a23872f
eac223c
 
 
 
 
 
 
 
 
 
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
eac223c
a23872f
eac223c
a23872f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eac223c
a23872f
 
 
 
 
 
 
 
 
 
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
# from functools import cache
import importlib

import gradio as gr
import matplotlib.pyplot as plt
import torch
import torchvision
import wandb
from icecream import ic
from torch import nn
from torchvision.transforms.functional import resize
from tqdm import tqdm
from transformers import CLIPModel, CLIPProcessor
import lpips
from edit import blend_paths
from img_processing import *
from img_processing import custom_to_pil
from loaders import load_default
import glob
import gc

global log
log=False

# ic.disable()
# ic.enable()
def get_resized_tensor(x):
    if len(x.shape) == 2:
        re = x.unsqueeze(0)
    else: re = x
    re = resize(re, (10, 10))
    return re
class ProcessorGradientFlow():
    """
    This wraps the huggingface CLIP processor to allow backprop through the image processing step.
    The original processor forces conversion to numpy then PIL images, which is faster for image processing but breaks gradient flow. 
    """
    def __init__(self, device="cuda") -> None:
        self.device = device
        self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
        self.image_mean = [0.48145466, 0.4578275, 0.40821073]
        self.image_std = [0.26862954, 0.26130258, 0.27577711]
        self.normalize = torchvision.transforms.Normalize(
            self.image_mean,
            self.image_std
        )
        self.resize = torchvision.transforms.Resize(224)
        self.center_crop = torchvision.transforms.CenterCrop(224)
    def preprocess_img(self, images):
        images = self.center_crop(images)
        images = self.resize(images)
        images = self.center_crop(images)
        images = self.normalize(images)
        return images
    def __call__(self, images=[], **kwargs):
        processed_inputs = self.processor(**kwargs)
        processed_inputs["pixel_values"] = self.preprocess_img(images)
        processed_inputs = {key:value.to(self.device) for (key, value) in processed_inputs.items()}
        return processed_inputs

class ImagePromptOptimizer(nn.Module):
    def __init__(self, 
                vqgan, 
                clip,
                clip_preprocessor,
                lpips_fn,
                iterations=100,
                lr = 0.01,
                save_vector=True,
                return_val="vector",
                quantize=True,
                make_grid=False,
                lpips_weight = 6.2) -> None:
                
        super().__init__()
        self.latent = None
        self.device = vqgan.device
        vqgan.eval()
        self.vqgan = vqgan
        self.clip = clip
        self.iterations = iterations
        self.lr = lr
        self.clip_preprocessor = clip_preprocessor
        self.make_grid = make_grid
        self.return_val = return_val
        self.quantize = quantize
        self.lpips_weight = lpips_weight
        self.perceptual_loss = lpips_fn
    def set_latent(self, latent):
        self.latent = latent.detach().to(self.device)
    def set_params(self, lr, iterations, lpips_weight, reconstruction_steps, attn_mask):
        self._attn_mask = attn_mask
        self.iterations = iterations
        self.lr = lr
        self.lpips_weight = lpips_weight
        self.reconstruction_steps = reconstruction_steps
    def forward(self, vector):
        base_latent = self.latent.detach().requires_grad_()
        trans_latent = base_latent + vector
        if self.quantize:
            z_q, *_ = self.vqgan.quantize(trans_latent)
        else:
            z_q = trans_latent
        dec = self.vqgan.decode(z_q)
        return dec
    def _get_clip_similarity(self, prompts, image, weights=None):
        if isinstance(prompts, str):
            prompts = [prompts]
        elif not isinstance(prompts, list):
            raise TypeError("Provide prompts as string or list of strings")
        clip_inputs = self.clip_preprocessor(text=prompts,
            images=image, return_tensors="pt", padding=True)
        clip_outputs = self.clip(**clip_inputs)
        similarity_logits = clip_outputs.logits_per_image
        if weights:
            similarity_logits *= weights
        return similarity_logits.sum()
    def get_similarity_loss(self, pos_prompts, neg_prompts, image):
        pos_logits = self._get_clip_similarity(pos_prompts, image)
        if neg_prompts:
            neg_logits = self._get_clip_similarity(neg_prompts, image)
        else:
            neg_logits = torch.tensor([1], device=self.device)
        loss = -torch.log(pos_logits) + torch.log(neg_logits)
        return loss
    def visualize(self, processed_img):
        if self.make_grid:
            self.index += 1
            plt.subplot(1, 13, self.index)
            plt.imshow(get_pil(processed_img[0]).detach().cpu())
        else:
            plt.imshow(get_pil(processed_img[0]).detach().cpu())
            plt.show()
    def _attn_mask(self, grad):
        newgrad = grad
        if self._attn_mask is not None:
            newgrad = grad * (self._attn_mask)
        return newgrad
    def _attn_mask_inverse(self, grad):
        newgrad = grad
        if self._attn_mask is not None:
            newgrad = grad * ((self._attn_mask - 1) * -1)
        return newgrad
    def _get_next_inputs(self, transformed_img):
        processed_img = loop_post_process(transformed_img) #* self.attn_mask
        processed_img.retain_grad()
        lpips_input = processed_img.clone()
        lpips_input.register_hook(self._attn_mask_inverse)
        lpips_input.retain_grad()
        clip_input = processed_img.clone()
        clip_input.register_hook(self._attn_mask)
        clip_input.retain_grad()
        return processed_img, lpips_input, clip_input

    def optimize(self, latent, pos_prompts, neg_prompts):
        self.set_latent(latent)
        transformed_img = self(torch.zeros_like(self.latent, requires_grad=True, device=self.device))
        original_img = loop_post_process(transformed_img)
        vector = torch.randn_like(self.latent, requires_grad=True, device=self.device)
        optim = torch.optim.Adam([vector], lr=self.lr)
        if self.make_grid:
            plt.figure(figsize=(35, 25))
            self.index = 1
        for i in tqdm(range(self.iterations)):
            optim.zero_grad()
            transformed_img = self(vector)
            processed_img, lpips_input, clip_input = self._get_next_inputs(transformed_img)
            with torch.autocast("cuda"):
                clip_loss = self.get_similarity_loss(pos_prompts, neg_prompts, clip_input)
                print("CLIP loss", clip_loss)
                perceptual_loss = self.perceptual_loss(lpips_input, original_img.clone()) * self.lpips_weight
                print("LPIPS loss: ", perceptual_loss)
            if log:
                wandb.log({"Perceptual Loss": perceptual_loss})
                wandb.log({"CLIP Loss": clip_loss})
            clip_loss.backward(retain_graph=True)
            perceptual_loss.backward(retain_graph=True)
            p2 = processed_img.grad
            print("Sum Loss", perceptual_loss + clip_loss)
            optim.step()
            # if i % self.iterations // 10 == 0: 
                # self.visualize(transformed_img)
            yield vector
        if self.make_grid:
            plt.savefig(f"plot {pos_prompts[0]}.png")
            plt.show()
        print("lpips solo op")
        for i in range(self.reconstruction_steps):
            optim.zero_grad()
            transformed_img = self(vector)
            processed_img = loop_post_process(transformed_img) #* self.attn_mask
            processed_img.retain_grad()
            lpips_input = processed_img.clone()
            lpips_input.register_hook(self._attn_mask_inverse)
            lpips_input.retain_grad()
            with torch.autocast("cuda"):
                perceptual_loss = self.perceptual_loss(lpips_input, original_img.clone()) * self.lpips_weight
            if log:
                wandb.log({"Perceptual Loss": perceptual_loss})
            print("LPIPS loss: ", perceptual_loss)
            perceptual_loss.backward(retain_graph=True)
            optim.step()
            yield vector
        yield vector if self.return_val == "vector" else self.latent + vector