Upload app.py

Added app.py file

app.py

@@ -0,0 +1,128 @@
+# -*- coding: utf-8 -*-
+"""Gradio with DocFormer
+
+Automatically generated by Colaboratory.
+
+Original file is located at
+    https://colab.research.google.com/drive/1_XBurG-8jYF4eJJK5VoCJ2Y1v6RV9iAW
+"""
+
+## Requirements.txt
+import os
+os.system('pip install pyyaml==5.1')
+## install PyTesseract
+os.system('pip install -q pytesseract')
+
+## Importing the functions from the DocFormer Repo
+from dataset import create_features
+from modeling import DocFormerEncoder,ResNetFeatureExtractor,DocFormerEmbeddings,LanguageFeatureExtractor
+from transformers import BertTokenizerFast
+from utils import DocFormer
+
+## Hyperparameters
+import torch
+
+seed = 42
+target_size = (500, 384)
+max_len = 128
+
+## Setting some hyperparameters
+
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+
+config = {
+  "coordinate_size": 96,              ## (768/8), 8 for each of the 8 coordinates of x, y
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "image_feature_pool_shape": [7, 7, 256],
+  "intermediate_ff_size_factor": 4,
+  "max_2d_position_embeddings": 1024,
+  "max_position_embeddings": 128,
+  "max_relative_positions": 8,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "shape_size": 96,
+  "vocab_size": 30522,
+  "layer_norm_eps": 1e-12,
+}
+
+## Defining the tokenizer
+tokenizer = BertTokenizerFast.from_pretrained("bert-base-uncased")
+
+
+
+docformer = DocFormer(config)
+
+# path_to_weights = 'drive/MyDrive/docformer_rvl_checkpoint/docformer_v1.ckpt'
+
+url = 'https://www.kaggleusercontent.com/kf/97691030/eyJhbGciOiJkaXIiLCJlbmMiOiJBMTI4Q0JDLUhTMjU2In0..64MVC5RwlflRqMaApK2jLw.rDiswzBHQcP_1_7vsHlJgSGKLdOqVB-d4hcGP6kQs5vEAdBmOzXL6XY9MleO3A4Sk0D5RB9QGeOyp7MuBZoHJbZ0gOVz6iRsats32fz2OU1yqQt22HIigL2mD_7mrTMn5IkP7KwsxtMMEuaOPEzFh1z8JQ9eE_NFBxIkOFF_Bp62a7agvDPL3HxzmxFQ7pwrYv9ZjYNfbDeeBuHu5J_MT_wHE5hOT1FENIMhebg3Q9l7eegUZD3eCMV4QoI_HsU6NZjyZOQcpVFmU6exYz8hGnFUa_V03870N6VnTkox78td0OXH29o3bYGSWneuCc86qSHKj5I1m8KbmCenPT6zU6IQINXp8BGLVlLOHdwVAPapR4X4CqSiK3Wgt5JINfpfVjQYWo2gDkAwJI026-fdLAfJQUI6mYGd-ERpyL5ZIbdkpesTslstOtlzoNT9gp_USW6aINxO8DranfK3-PiMZ_X1zHsK1vscRpO9gohNhuOg362ijjl3FQrw48-YbYfykQFfVwQpnhYQ9Q6d5gNANfJMrzH92DlpQFBaPOLcze1BAVdM4zmVGdt8Jo-Knk1JADpNizHWmF19eDxudQO_ZCxvXWpc8v3LOh-HpA2mBB0HI1DZ4cqcMETtOwas5wzHrLqDLRJpso6BKOgz78kIZJDdj6rr7yY4QVWpVOOdNZ8.VZzPPNhnz_MUdNnc5DaZOw/models/epoch=0-step=753.ckpt'
+
+docformer.load_from_checkpoint(url)
+
+id2label = ['scientific_report',
+ 'resume',
+ 'memo',
+ 'file_folder',
+ 'specification',
+ 'news_article',
+ 'letter',
+ 'form',
+ 'budget',
+ 'handwritten',
+ 'email',
+ 'invoice',
+ 'presentation',
+ 'scientific_publication',
+ 'questionnaire',
+ 'advertisement']
+
+import gradio as gr
+
+## Taken from LayoutLMV2 space
+
+image = gr.inputs.Image(type="pil")
+label = gr.outputs.Label(num_top_classes=5)
+examples = [['00093726.png'], ['00866042.png']]
+title = "Interactive demo: DocFormer for Image Classification"
+description = "Demo for classifying document images with DocFormer model. To use it, \
+simply upload an image or use the example images below and click 'submit' to let the model predict the 5 most probable Document classes. \
+Results will show up in a few seconds."
+
+def classify_image(image):
+
+  image.save('sample_img.png')
+  final_encoding = create_features(
+            './sample_img.png',
+            tokenizer,
+            add_batch_dim=True,
+            target_size=target_size,
+            max_seq_length=max_len,
+            path_to_save=None,
+            save_to_disk=False,
+            apply_mask_for_mlm=False,
+            extras_for_debugging=False,
+            use_ocr = True
+    )
+
+  keys_to_reshape = ['x_features', 'y_features', 'resized_and_aligned_bounding_boxes']
+  for key in keys_to_reshape:
+      final_encoding[key] = final_encoding[key][:, :max_len]
+
+  from torchvision import transforms
+  # ## Normalization to these mean and std (I have seen some tutorials used this, and also in image reconstruction, so used it)
+  transform = transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
+
+  final_encoding['resized_scaled_img'] = transform(final_encoding['resized_scaled_img'])
+  output = docformer.forward(final_encoding)
+  output = output[0].softmax(axis = -1)
+  
+  final_pred = {}
+  for i, score in enumerate(output):
+      score = output[i]
+      final_pred[id2label[i]] = score.detach().cpu().tolist()
+      
+  return final_pred
+
+gr.Interface(fn=classify_image, inputs=image, outputs=label, title=title, description=description, examples=examples, enable_queue=True).launch(debug=True)
+