AnoPILaD/rec_generate.py at main · QuIIL/AnoPILaD · GitHub

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
import sys
sys.path.append('./')
sys.path.append('../')
sys.path.append('../../')

import argparse
from pathlib import Path
import os
import csv
import torch
from tqdm import tqdm
from torch.utils.data import Dataset
from eval_sampler import DistributedEvalSampler
os.environ['CUDA_VISIBLE_DEVICES']='0,1'
from munch import munchify
from torchvision.utils import save_image
from PIL import Image
from ldm_solvers import get_solver
from utils import create_workdir, set_seed
import pandas as pd
import numpy as np
import lpips
import json
from accelerate import PartialState
from torchvision import transforms


class RecDataset(Dataset):
    def __init__(self, csv_path, image_root_path, transformer, out_dir):
        self.csv_path = csv_path
        self.image_root_path = image_root_path
        self.transformer = transformer
        csv_file = pd.read_csv(csv_path)
        print(f'Load for {len(list(csv_file["file_name"]))} samples')

        done_fn_list = []
        for path, dir, files in os.walk(os.path.join(out_dir, "result")):
            for fn in files:
                ext = os.path.splitext(fn)[-1]
                if ext.lower() not in ('.png', '.jpg'):
                    continue
                file_path = os.path.join(path.split('/')[-1], fn)
                done_fn_list.append(file_path)

        done_fn_list = [f"{fn.split('.')[0][:-4]}.png" for fn in done_fn_list]
        csv_file = csv_file.loc[~csv_file["file_name"].isin(done_fn_list)]
        self.file_names = list(csv_file["file_name"])
        self.texts = list(csv_file["text"])
        print(f'Again Load for {len(list(csv_file["file_name"]))} samples')

    def __len__(self):
        return len(self.file_names)

    def __getitem__(self, index):
        fn = self.file_names[index]
        text = self.texts[index]
        image_path = os.path.join(self.image_root_path, fn)
        image = Image.open(image_path).convert("RGB")
        image = self.transformer(image)

        dir = fn.split('/')[0]
        fn = fn.split('/')[1]
        basename = fn.split('.')[0]
        return dir, fn, basename, image, text

def main():
    file = 'config.json'
    with open(file, 'r') as f:
        f_args = json.load(f)

    parser = argparse.ArgumentParser(description="AnoPILAD")
    parser.add_argument("--workdir", type=Path, default=f_args["workdir"])
    parser.add_argument("--dir_type", type=str, default=f_args["dir_type"])
    parser.add_argument("--pipeline_path", type=str, default=f_args["pipeline_path"])
    parser.add_argument("--data_path", type=str, default=f_args["data_path"])
    parser.add_argument("--NFE", type=int, default=f_args["NFE"])
    parser.add_argument("--start_lambda", type=int, default=f_args["start_lambda"])
    parser.add_argument("--img_resolution", type=Path, default=f_args["img_resolution"])
    parser.add_argument("--num_workers", type=str, default=f_args["num_workers"])
    parser.add_argument("--batch_size", type=str, default=f_args["batch_size"])
    parser.add_argument("--seed", type=int, default=88888888)
    parser.add_argument("--cfg_guidance", type=float, default=7.5)
    parser.add_argument("--method", type=str, default="ddim")

    args = parser.parse_args()
    set_seed(args.seed)
    distributed_state = PartialState()
    solver_config = munchify({'num_sampling': args.NFE})
    solver = get_solver(args.method,
                        pipeline_path=args.pipeline_path,
                        solver_config=solver_config,
                        device=distributed_state.device)

    csv_path = os.path.join(args.data_path, "metadata.csv")
    fieldnames = ["file_name", "mse_latent", "lpips_pixel"]
    output_loss_path = os.path.join(args.workdir, args.dir_type, str(args.start_lambda), 'loss.csv')
    output_rec_path = os.path.join(args.workdir, args.dir_type, str(args.start_lambda))
    create_workdir(output_rec_path, "image_result")

    if not os.path.isfile(output_loss_path):
        with open(output_loss_path, mode='w', newline='') as file:
            writer = csv.DictWriter(file, fieldnames=fieldnames)
            # Write the header
            writer.writeheader()

    img_transforms = transforms.Compose(
        [
            transforms.Resize(args.img_resolution),
            transforms.ToTensor(),
            transforms.Normalize([0.5], [0.5]),
        ]
    )
    test_dataset = RecDataset(csv_path, args.data_path, img_transforms, output_rec_path)
    sampler = DistributedEvalSampler(test_dataset, rank=distributed_state.process_index, num_replicas=distributed_state.num_processes)
    test_dataloader = torch.utils.data.DataLoader(test_dataset, batch_size=args.batch_size,
                                                  num_workers=args.num_workers, shuffle=False,
                                                  sampler=sampler
                                                  )
    loss_lpips = lpips.LPIPS(net='vgg').to(distributed_state.device)
    for i, (subfolder, fn, basename, src_img, prompt) in tqdm(enumerate(test_dataloader), total=len(test_dataloader)):
        src_img = src_img.to(distributed_state.device)
        with torch.no_grad():
            result, loss = solver.sample_forward_backward(gt=src_img,
                                                             prompt=[list(np.repeat("", len(subfolder))),
                                                                     list(prompt)],
                                                             cfg_guidance=args.cfg_guidance,
                                                             target_size=(256, 256),
                                                             start_lambda=args.start_lambda,
                                                             gamma=1e-1)
            lpips_pixel = loss_lpips(src_img, result).squeeze(1, 2, 3)
            result = result / 2 + 0.5
        write_blocks = []
        for j in range(len(subfolder)):
            os.makedirs(os.path.join(output_rec_path, 'image_result', subfolder[j]), exist_ok=True)
            save_image(result[j], f'{output_rec_path}/image_result/{subfolder[j]}/{basename[j]}_rec.png', normalize=True)
            write_blocks.append({
                "file_name": fn[j],
                "mse_latent": str(loss[j].item()),
                "lpips_pixel": str(lpips_pixel[j].item()),
            })
        with open(output_loss_path, mode='a', newline='') as file:
            writer = csv.DictWriter(file, fieldnames=fieldnames)
            writer.writerows(write_blocks)


if __name__ == "__main__":

    main()