update space

.gitignore

@@ -0,0 +1,2 @@
+/.env
+/dummy.py

app.py

@@ -1,14 +1,26 @@
+import time
 import os
-os.system('pip install detectron2@git+https://github.com/facebookresearch/detectron2.git')
+from os import getcwd, path
+import importlib.metadata
+from dotenv import load_dotenv
 
-# work around: https://discuss.huggingface.co/t/how-to-install-a-specific-version-of-gradio-in-spaces/13552
-os.system("pip uninstall -y gradio")
-os.system("pip install gradio==3.4.1")
-os.system("pip install packaging==21.3")
-os.system(os.environ["DD_ADDONS"])
 
-import time
-from os import getcwd, path
+def check_additional_requirements():
+    if importlib.util.find_spec("detectron2") is None:
+        os.system('pip install detectron2@git+https://github.com/facebookresearch/detectron2.git')
+    if importlib.util.find_spec("gradio") is not None:
+        if importlib.metadata.version("gradio")!="3.4.1":
+            os.system("pip uninstall -y gradio")
+            os.system("pip install gradio==3.4.1")
+    else:
+        os.system("pip install gradio==3.4.1")
+    os.system(os.environ["DD_ADDONS"])
+    return
+
+
+load_dotenv()
+check_additional_requirements()
+
 
 import deepdoctection as dd
 from deepdoctection.dataflow.serialize import DataFromList
@@ -21,7 +33,10 @@ from dd_addons.extern.openai import OpenAiLmmTokenClassifier, is_api_key_valid
 
 import gradio as gr
 
-analyzer = get_loader(reset_config_file=True, config_overwrite=["OCR.USE_TESSERACT=False","OCR.USE_TEXTRACT=True"])
+dd.Page.add_attribute_name("raw_json_output")
+analyzer = get_loader(reset_config_file=True, config_overwrite=["OCR.USE_TESSERACT=False",
+                                                                "OCR.USE_TEXTRACT=True",
+                                                                "WORD_MATCHING.MAX_PARENT_ONLY=True"])
 
 demo = gr.Blocks(css="scrollbar.css")
 
@@ -30,6 +45,9 @@ def process_analyzer(openai_api_key, categories_str, instruction_str, img, pdf,
     if not is_api_key_valid(openai_api_key):
         return [], {}, "You have entered no or an invalid api key. Please enter a valid api key"
     categories_list = categories_str.split(",")
+    if not categories_str:
+        return [], {}, "You did not enter any entities. Please enter a at least one category."
+
     register_string_categories_from_list(categories_list, "custom_token_classes")
     custom_token_class = dd.object_types_registry.get("custom_token_classes")
     print([token_class for token_class in custom_token_class])
@@ -61,13 +79,15 @@ def process_analyzer(openai_api_key, categories_str, instruction_str, img, pdf,
 
     json_out = {}
     dpts = []
+    json_out_raw = {}
 
     for idx, dp in enumerate(df):
         dpts.append(dp)
         json_out[f"page_{idx}"] = dp.get_token()
+        json_out_raw[f"page_{idx}"] = dp.raw_json_output
 
     return [dp.viz(show_cells=False, show_layouts=False, show_tables=False, show_words=True, show_token_class=True, ignore_default_token_class=True)
-            for dp in dpts], json_out, "No error"
+            for dp in dpts], json_out, json_out_raw, "No error"
 
 
 with demo:
@@ -125,18 +145,20 @@ with demo:
                 with gr.Box():
                     gr.Markdown("<center><strong>JSON</strong></center>")
                     json = gr.JSON()
+                with gr.Box():
+                    gr.Markdown("<center><strong>ChatGPT output. </strong> <br />"
+                                "It is possible that ChatGPT answers in an unexpected way, "
+                                "such that the answer cannot be properly processed. In this case you might get"
+                                "an empty JSON but you can still see the raw output.</center>")
+                    json_raw = gr.JSON()
             with gr.Column():
                 with gr.Box():
                     gr.Markdown("<center><strong>Layout detection</strong></center>")
                     gallery = gr.Gallery(
                         label="Output images", show_label=False, elem_id="gallery"
                     ).style(grid=2)
-        with gr.Row():
-            with gr.Box():
-                gr.Markdown("<center><strong>Table</strong></center>")
-                html = gr.HTML()
 
     btn.click(fn=process_analyzer, inputs=[user_token, categories,  instruction, inputs, inputs_pdf, max_imgs],
-              outputs=[gallery, json, msg])
+              outputs=[gallery, json, json_raw, msg])
 
 demo.launch()

requirements.txt

@@ -1,3 +1,4 @@
+python-dotenv
 Pillow==9.5.0
 torch==1.12.0
 torchvision==0.13.0