forked from facebookresearch/nougat
-
Notifications
You must be signed in to change notification settings - Fork 0
/
predict_imgs.py
75 lines (55 loc) · 2.83 KB
/
predict_imgs.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
import torch
from torch.utils.data import Dataset, DataLoader
from PIL import Image
from functools import partial
from nougat import NougatModel
import argparse
import os
import time
class ImageDataset(Dataset):
def __init__(self, image_paths, preprocess_fn):
self.image_paths = image_paths
self.preprocess_fn = preprocess_fn
def __len__(self):
return len(self.image_paths)
def __getitem__(self, idx):
image = Image.open(self.image_paths[idx])
target_width = 876
original_width, original_height = image.size
new_height = int((target_width / original_width) * original_height)
#image = image.resize((target_width, new_height), Image.Resampling.LANCZOS)
image_tensor = self.preprocess_fn(image)
return image_tensor, os.path.basename(self.image_paths[idx])
def main(model_path, input_dir, output_dir, batch_size):
model = NougatModel.from_pretrained(model_path)
model.eval()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
preprocess_fn = partial(model.encoder.prepare_input, random_padding=False)
image_paths = [os.path.join(input_dir, fname) for fname in os.listdir(input_dir) if fname.endswith('.png')]
image_dataset = ImageDataset(image_paths, preprocess_fn)
dataloader = DataLoader(image_dataset, batch_size=batch_size, shuffle=False)
os.makedirs(output_dir, exist_ok=True)
total_inference_time = 0
for batch, filenames in dataloader:
batch = batch.to(device)
start_time = time.time()
output = model.inference(image_tensors=batch, early_stopping=False)
inference_time = time.time() - start_time
total_inference_time += inference_time
predictions = output['predictions']
for filename, prediction in zip(filenames, predictions):
print(prediction)
output_path = os.path.join(output_dir, filename.replace('.png', '.txt'))
with open(output_path, 'w') as f:
f.write(prediction)
print(f"Total inference time: {total_inference_time:.2f} seconds")
print(f"Average inference time per image: {total_inference_time / len(image_paths):.2f} seconds")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Batch infer images with NougatModel.")
parser.add_argument('--model_path', type=str, required=True, help="Path to the pretrained model.")
parser.add_argument('--input_dir', type=str, required=True, help="Directory containing input images.")
parser.add_argument('--output_dir', type=str, required=True, help="Directory to save output predictions.")
parser.add_argument('--batch_size', type=int, default=4, help="Batch size for inference.")
args = parser.parse_args()
main(args.model_path, args.input_dir, args.output_dir, args.batch_size)