Spaces:
Runtime error
Runtime error
first-commit
Browse files- classifier_app.py +36 -0
classifier_app.py
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
def classify(input_img):
|
| 5 |
+
|
| 6 |
+
from transformers import (
|
| 7 |
+
AutoModelForSequenceClassification,
|
| 8 |
+
LayoutLMv2FeatureExtractor,
|
| 9 |
+
LayoutLMv2Tokenizer,
|
| 10 |
+
LayoutLMv2Processor,
|
| 11 |
+
)
|
| 12 |
+
|
| 13 |
+
model = AutoModelForSequenceClassification.from_pretrained(
|
| 14 |
+
"fedihch/InvoiceReceiptClassifier"
|
| 15 |
+
)
|
| 16 |
+
feature_extractor = LayoutLMv2FeatureExtractor()
|
| 17 |
+
tokenizer = LayoutLMv2Tokenizer.from_pretrained("microsoft/layoutlmv2-base-uncased")
|
| 18 |
+
processor = LayoutLMv2Processor(feature_extractor, tokenizer)
|
| 19 |
+
encoded_inputs = processor(input_img, return_tensors="pt")
|
| 20 |
+
for k, v in encoded_inputs.items():
|
| 21 |
+
encoded_inputs[k] = v.to(model.device)
|
| 22 |
+
outputs = model(**encoded_inputs)
|
| 23 |
+
logits = outputs.logits
|
| 24 |
+
predicted_class_idx = logits.argmax(-1).item()
|
| 25 |
+
|
| 26 |
+
id2label = {0: "invoice", 1: "receipt"}
|
| 27 |
+
return id2label[predicted_class_idx]
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
demo = gr.Interface(
|
| 31 |
+
fn=classify,
|
| 32 |
+
inputs=gr.Image(shape=(200, 200)),
|
| 33 |
+
outputs="text",
|
| 34 |
+
allow_flagging="manual",
|
| 35 |
+
)
|
| 36 |
+
demo.launch(share=True)
|