stupidog04
commited on
Commit
•
151724b
1
Parent(s):
45c38e5
Update pipeline.py
Browse files- pipeline.py +27 -0
pipeline.py
CHANGED
@@ -1,8 +1,35 @@
|
|
1 |
from torchvision import transforms
|
|
|
2 |
from transformers import ImageClassificationPipeline
|
3 |
import torch
|
4 |
|
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
class PairClassificationPipeline(ImageClassificationPipeline):
|
7 |
pipe_to_tensor = transforms.ToTensor()
|
8 |
pipe_to_pil = transforms.ToPILImage()
|
|
|
1 |
from torchvision import transforms
|
2 |
+
from transformers import ViTFeatureExtractor, ViTForImageClassification
|
3 |
from transformers import ImageClassificationPipeline
|
4 |
import torch
|
5 |
|
6 |
|
7 |
+
class PreTrainedPipeline():
|
8 |
+
def __init__(self, path):
|
9 |
+
"""
|
10 |
+
Initialize model
|
11 |
+
"""
|
12 |
+
# self.processor = feature_extractor = ViTFeatureExtractor.from_pretrained(model_flag)
|
13 |
+
model_flag = 'google/vit-base-patch16-224-in21k'
|
14 |
+
# model_flag = 'google/vit-base-patch16-384'
|
15 |
+
self.feature_extractor = ViTFeatureExtractor.from_pretrained(model_flag)
|
16 |
+
self.model = ViTForImageClassification.from_pretrained(path)
|
17 |
+
self.pipe = PairClassificationPipeline(self.model, feature_extractor=self.feature_extractor)
|
18 |
+
|
19 |
+
def __call__(self, inputs):
|
20 |
+
"""
|
21 |
+
Args:
|
22 |
+
inputs (:obj:`np.array`):
|
23 |
+
The raw waveform of audio received. By default at 16KHz.
|
24 |
+
Return:
|
25 |
+
A :obj:`dict`:. The object return should be liked {"text": "XXX"} containing
|
26 |
+
the detected text from the input audio.
|
27 |
+
"""
|
28 |
+
# input_values = self.processor(inputs, return_tensors="pt", sampling_rate=self.sampling_rate).input_values # Batch size 1
|
29 |
+
# logits = self.model(input_values).logits.cpu().detach().numpy()[0]
|
30 |
+
return self.pipe(inputs)
|
31 |
+
|
32 |
+
|
33 |
class PairClassificationPipeline(ImageClassificationPipeline):
|
34 |
pipe_to_tensor = transforms.ToTensor()
|
35 |
pipe_to_pil = transforms.ToPILImage()
|