Upload processor
Browse files- image_processing_resnet.py +66 -0
- preprocessor_config.json +25 -0
image_processing_resnet.py
ADDED
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
|
3 |
+
import albumentations as A
|
4 |
+
from albumentations.pytorch.transforms import ToTensorV2
|
5 |
+
import PIL.Image
|
6 |
+
import numpy as np
|
7 |
+
from functools import partial
|
8 |
+
from typing import Dict, List, Optional, Union
|
9 |
+
from datasets import load_dataset, DatasetDict, Image # pip install datasets
|
10 |
+
from torch.utils.data import DataLoader
|
11 |
+
import torch
|
12 |
+
|
13 |
+
# 自定义 ImageProcessor 为了与 pipeline使用
|
14 |
+
from transformers import ViTImageProcessor
|
15 |
+
from transformers.image_utils import PILImageResampling, ChannelDimension
|
16 |
+
from transformers.image_processing_utils import get_size_dict
|
17 |
+
|
18 |
+
class ResnetImageProcessor(ViTImageProcessor):
|
19 |
+
"""
|
20 |
+
>>> # tfs = A.Compose([A.Resize(256, 256), A.CenterCrop(224, 224)])
|
21 |
+
>>> # 如果传入 参数 tfs=tfs 在调用save_pretrained会报错
|
22 |
+
>>> # 本地使用
|
23 |
+
>>> mean = [0.485, 0.456, 0.406];std = [0.229, 0.224, 0.225]
|
24 |
+
>>> image_processor = ResnetImageProcessor(size=(224, 224), image_mean=mean, image_std=std)
|
25 |
+
>>> image_processor.save_pretrained("custom-resnet")
|
26 |
+
>>> image_processor = ResnetImageProcessor.from_pretrained("custom-resnet")
|
27 |
+
|
28 |
+
>>> # push_to_hub
|
29 |
+
>>> # hub登录
|
30 |
+
>>> from huggingface_hub import notebook_login;notebook_login()
|
31 |
+
>>> # or huggingface-cli login
|
32 |
+
|
33 |
+
>>> ResnetImageProcessor.register_for_auto_class()
|
34 |
+
>>> mean = [0.485, 0.456, 0.406];std = [0.229, 0.224, 0.225]
|
35 |
+
>>> image_processor = ResnetImageProcessor(size=(224, 224), image_mean=mean, image_std=std)
|
36 |
+
>>> image_processor.save_pretrained("custom-resnet")
|
37 |
+
>>> # image_processor = ResnetImageProcessor.from_pretrained("custom-resnet")
|
38 |
+
>>> # 如果要执行 push_to_hub 需要将 custom-resnet/preprocessor_config.json 中的 "image_processor_type" 改成 "ViTImageProcessor"
|
39 |
+
>>> # 默认的 ResnetImageProcessor 没有注册到 AutoImageProcessor
|
40 |
+
>>> # 否则从 使用 AutoImageProcessor 加载 会报错了
|
41 |
+
>>> image_processor.push_to_hub('custom-resnet')
|
42 |
+
|
43 |
+
>>> # 从 huggingface hub 加载
|
44 |
+
>>> from transformers import AutoImageProcessor
|
45 |
+
>>> AutoImageProcessor.register(config_class='wucng/custom-resnet/config.json',image_processor_class=ResnetImageProcessor)
|
46 |
+
>>> image_processor = AutoImageProcessor.from_pretrained('wucng/custom-resnet', trust_remote_code=True)
|
47 |
+
"""
|
48 |
+
|
49 |
+
def resize(
|
50 |
+
self,
|
51 |
+
image: np.ndarray,
|
52 |
+
size: Dict[str, int],
|
53 |
+
resample: PILImageResampling = PILImageResampling.BILINEAR,
|
54 |
+
data_format: Optional[Union[str, ChannelDimension]] = None,
|
55 |
+
input_data_format: Optional[Union[str, ChannelDimension]] = None,
|
56 |
+
**kwargs,
|
57 |
+
) -> np.ndarray:
|
58 |
+
size = get_size_dict(size)
|
59 |
+
output_size = (size["height"], size["width"])
|
60 |
+
height, width = size["height"], size["width"]
|
61 |
+
|
62 |
+
tfs = kwargs.get('tfs', None)
|
63 |
+
if tfs is None:
|
64 |
+
ratio = 256 / 224
|
65 |
+
tfs = A.Compose([A.Resize(int(ratio * height), int(ratio * width)), A.CenterCrop(height, width)])
|
66 |
+
return tfs(image=image)['image']
|
preprocessor_config.json
ADDED
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"auto_map": {
|
3 |
+
"AutoImageProcessor": "image_processing_resnet.ResnetImageProcessor"
|
4 |
+
},
|
5 |
+
"do_normalize": true,
|
6 |
+
"do_rescale": true,
|
7 |
+
"do_resize": true,
|
8 |
+
"image_mean": [
|
9 |
+
0.485,
|
10 |
+
0.456,
|
11 |
+
0.406
|
12 |
+
],
|
13 |
+
"image_processor_type": "ResnetImageProcessor",
|
14 |
+
"image_std": [
|
15 |
+
0.229,
|
16 |
+
0.224,
|
17 |
+
0.225
|
18 |
+
],
|
19 |
+
"resample": 2,
|
20 |
+
"rescale_factor": 0.00392156862745098,
|
21 |
+
"size": {
|
22 |
+
"height": 224,
|
23 |
+
"width": 224
|
24 |
+
}
|
25 |
+
}
|