docling-ibm-models 1.2.0__tar.gz → 1.3.0__tar.gz

{docling_ibm_models-1.2.0 → docling_ibm_models-1.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: docling-ibm-models
-Version: 1.2.0
+Version: 1.3.0
 Summary: This package contains the AI models used by the Docling PDF conversion package
 License: MIT
 Keywords: docling,convert,document,pdf,layout model,segmentation,table structure,table former
@@ -19,6 +19,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Dist: Pillow (>=10.0.0,<11.0.0)
+Requires-Dist: huggingface_hub (>=0.23,<1)
 Requires-Dist: jsonlines (>=3.1.0,<4.0.0)
 Requires-Dist: lxml (>=4.9.1,<5.0.0)
 Requires-Dist: mean_average_precision (>=2021.4.26.0,<2022.0.0.0)
@@ -110,7 +111,7 @@ Below we list datasets used with their description, source, and ***"TableFormer
 
 ## Configuration file
 
-Example configuration can be seen inside test `tests/test_tf_predictor.py`
+Example configuration can be found inside test `tests/test_tf_predictor.py`
 These are the main sections of the configuration file:
 
 - `dataset`: The directory for prepared data and the parameters used during the data loading.
@@ -128,16 +129,13 @@ You can download the model weights and config files from the links:
 - [TableFormer Checkpoint](https://huggingface.co/ds4sd/docling-models/tree/main/model_artifacts/tableformer)
 - [beehive_v0.0.5](https://huggingface.co/ds4sd/docling-models/tree/main/model_artifacts/layout/beehive_v0.0.5)
 
-Place the downloaded files into `tests/test_data/model_artifacts/` directory.
-
 
 ## Inference Tests
 
-This contains unit tests for Docling models.
+You can run the inference tests for the models with:
 
-First download the model weights (see above), then run:
 ```
-./devtools/check_code.sh
+python -m pytest tests/
 ```
 
 This will also generate prediction and matching visualizations that can be found here:

{docling_ibm_models-1.2.0 → docling_ibm_models-1.3.0}/README.md

@@ -76,7 +76,7 @@ Below we list datasets used with their description, source, and ***"TableFormer
 
 ## Configuration file
 
-Example configuration can be seen inside test `tests/test_tf_predictor.py`
+Example configuration can be found inside test `tests/test_tf_predictor.py`
 These are the main sections of the configuration file:
 
 - `dataset`: The directory for prepared data and the parameters used during the data loading.
@@ -94,16 +94,13 @@ You can download the model weights and config files from the links:
 - [TableFormer Checkpoint](https://huggingface.co/ds4sd/docling-models/tree/main/model_artifacts/tableformer)
 - [beehive_v0.0.5](https://huggingface.co/ds4sd/docling-models/tree/main/model_artifacts/layout/beehive_v0.0.5)
 
-Place the downloaded files into `tests/test_data/model_artifacts/` directory.
-
 
 ## Inference Tests
 
-This contains unit tests for Docling models.
+You can run the inference tests for the models with:
 
-First download the model weights (see above), then run:
 ```
-./devtools/check_code.sh
+python -m pytest tests/
 ```
 
 This will also generate prediction and matching visualizations that can be found here:

{docling_ibm_models-1.2.0 → docling_ibm_models-1.3.0}/docling_ibm_models/tableformer/data_management/tf_cell_matcher.py

@@ -129,12 +129,15 @@ class CellMatcher:
         pdf_cells = copy.deepcopy(iocr_page["tokens"])
         if len(pdf_cells) > 0:
             for word in pdf_cells:
-                word["bbox"] = [
-                    word["bbox"]["l"],
-                    word["bbox"]["t"],
-                    word["bbox"]["r"],
-                    word["bbox"]["b"],
-                ]
+                if isinstance(word["bbox"], list):
+                    continue
+                elif isinstance(word["bbox"], dict):
+                    word["bbox"] = [
+                        word["bbox"]["l"],
+                        word["bbox"]["t"],
+                        word["bbox"]["r"],
+                        word["bbox"]["b"],
+                    ]
         table_bboxes = prediction["bboxes"]
         table_classes = prediction["classes"]
         # BBOXES transformed...

{docling_ibm_models-1.2.0 → docling_ibm_models-1.3.0}/docling_ibm_models/tableformer/data_management/tf_predictor.py

@@ -524,7 +524,12 @@ class TFPredictor:
         return resized, sf
 
     def multi_table_predict(
-        self, iocr_page, table_bboxes, do_matching=True, correct_overlapping_cells=False
+        self,
+        iocr_page,
+        table_bboxes,
+        do_matching=True,
+        correct_overlapping_cells=False,
+        sort_row_col_indexes=True,
     ):
         multi_tf_output = []
         page_image = iocr_page["image"]
@@ -563,56 +568,70 @@ class TFPredictor:
             # PROCESS PREDICTED RESULTS, TO TURN PREDICTED COL/ROW IDs into Indexes
             # Indexes should be in increasing order, without gaps
 
-            # Fix col/row indexes
-            # Arranges all col/row indexes sequentially without gaps using input IDs
-
-            indexing_start_cols = []  # Index of original start col IDs (not indexes)
-            indexing_end_cols = []  # Index of original end col IDs (not indexes)
-            indexing_start_rows = []  # Index of original start row IDs (not indexes)
-            indexing_end_rows = []  # Index of original end row IDs (not indexes)
-
-            # First, collect all possible predicted IDs, to be used as indexes
-            # ID's returned by Tableformer are sequential, but might contain gaps
-            for tf_response_cell in tf_responses:
-                start_col_offset_idx = tf_response_cell["start_col_offset_idx"]
-                end_col_offset_idx = tf_response_cell["end_col_offset_idx"]
-                start_row_offset_idx = tf_response_cell["start_row_offset_idx"]
-                end_row_offset_idx = tf_response_cell["end_row_offset_idx"]
-
-                # Collect all possible col/row IDs:
-                if start_col_offset_idx not in indexing_start_cols:
-                    indexing_start_cols.append(start_col_offset_idx)
-                if end_col_offset_idx not in indexing_end_cols:
-                    indexing_end_cols.append(end_col_offset_idx)
-                if start_row_offset_idx not in indexing_start_rows:
-                    indexing_start_rows.append(start_row_offset_idx)
-                if end_row_offset_idx not in indexing_end_rows:
-                    indexing_end_rows.append(end_row_offset_idx)
-
-            indexing_start_cols.sort()
-            indexing_end_cols.sort()
-            indexing_start_rows.sort()
-            indexing_end_rows.sort()
-
-            # After this - put actual indexes of IDs back into predicted structure...
-            for tf_response_cell in tf_responses:
-                tf_response_cell["start_col_offset_idx"] = indexing_start_cols.index(
-                    tf_response_cell["start_col_offset_idx"]
-                )
-                tf_response_cell["end_col_offset_idx"] = (
-                    tf_response_cell["start_col_offset_idx"]
-                    + tf_response_cell["col_span"]
-                )
-                tf_response_cell["start_row_offset_idx"] = indexing_start_rows.index(
-                    tf_response_cell["start_row_offset_idx"]
-                )
-                tf_response_cell["end_row_offset_idx"] = (
-                    tf_response_cell["start_row_offset_idx"]
-                    + tf_response_cell["row_span"]
-                )
-            # Counting matched cols/rows from actual indexes (and not ids)
-            predict_details["num_cols"] = len(indexing_end_cols)
-            predict_details["num_rows"] = len(indexing_end_rows)
+            if sort_row_col_indexes:
+                # Fix col/row indexes
+                # Arranges all col/row indexes sequentially without gaps using input IDs
+
+                indexing_start_cols = (
+                    []
+                )  # Index of original start col IDs (not indexes)
+                indexing_end_cols = []  # Index of original end col IDs (not indexes)
+                indexing_start_rows = (
+                    []
+                )  # Index of original start row IDs (not indexes)
+                indexing_end_rows = []  # Index of original end row IDs (not indexes)
+
+                # First, collect all possible predicted IDs, to be used as indexes
+                # ID's returned by Tableformer are sequential, but might contain gaps
+                for tf_response_cell in tf_responses:
+                    start_col_offset_idx = tf_response_cell["start_col_offset_idx"]
+                    end_col_offset_idx = tf_response_cell["end_col_offset_idx"]
+                    start_row_offset_idx = tf_response_cell["start_row_offset_idx"]
+                    end_row_offset_idx = tf_response_cell["end_row_offset_idx"]
+
+                    # Collect all possible col/row IDs:
+                    if start_col_offset_idx not in indexing_start_cols:
+                        indexing_start_cols.append(start_col_offset_idx)
+                    if end_col_offset_idx not in indexing_end_cols:
+                        indexing_end_cols.append(end_col_offset_idx)
+                    if start_row_offset_idx not in indexing_start_rows:
+                        indexing_start_rows.append(start_row_offset_idx)
+                    if end_row_offset_idx not in indexing_end_rows:
+                        indexing_end_rows.append(end_row_offset_idx)
+
+                indexing_start_cols.sort()
+                indexing_end_cols.sort()
+                indexing_start_rows.sort()
+                indexing_end_rows.sort()
+
+                # After this - put actual indexes of IDs back into predicted structure...
+                for tf_response_cell in tf_responses:
+                    tf_response_cell["start_col_offset_idx"] = (
+                        indexing_start_cols.index(
+                            tf_response_cell["start_col_offset_idx"]
+                        )
+                    )
+                    tf_response_cell["end_col_offset_idx"] = (
+                        tf_response_cell["start_col_offset_idx"]
+                        + tf_response_cell["col_span"]
+                    )
+                    tf_response_cell["start_row_offset_idx"] = (
+                        indexing_start_rows.index(
+                            tf_response_cell["start_row_offset_idx"]
+                        )
+                    )
+                    tf_response_cell["end_row_offset_idx"] = (
+                        tf_response_cell["start_row_offset_idx"]
+                        + tf_response_cell["row_span"]
+                    )
+                # Counting matched cols/rows from actual indexes (and not ids)
+                predict_details["num_cols"] = len(indexing_end_cols)
+                predict_details["num_rows"] = len(indexing_end_rows)
+            else:
+                otsl_seq = predict_details["prediction"]["rs_seq"]
+                predict_details["num_cols"] = otsl_seq.index("nl")
+                predict_details["num_rows"] = otsl_seq.count("nl")
+
             # Put results into multi_tf_output
             multi_tf_output.append(
                 {"tf_responses": tf_responses, "predict_details": predict_details}
@@ -667,13 +686,20 @@ class TFPredictor:
                 )
 
                 if outputs_coord is not None:
-                    bbox_pred = u.box_cxcywh_to_xyxy(outputs_coord)
-                    prediction["bboxes"] = bbox_pred.tolist()
+                    if len(outputs_coord) == 0:
+                        prediction["bboxes"] = []
+                    else:
+                        bbox_pred = u.box_cxcywh_to_xyxy(outputs_coord)
+                        prediction["bboxes"] = bbox_pred.tolist()
                 else:
                     prediction["bboxes"] = []
+
                 if outputs_class is not None:
-                    result_class = torch.argmax(outputs_class, dim=1)
-                    prediction["classes"] = result_class.tolist()
+                    if len(outputs_class) == 0:
+                        prediction["classes"] = []
+                    else:
+                        result_class = torch.argmax(outputs_class, dim=1)
+                        prediction["classes"] = result_class.tolist()
                 else:
                     prediction["classes"] = []
                 if self._remove_padding:
@@ -788,13 +814,20 @@ class TFPredictor:
                 )
 
                 if outputs_coord is not None:
-                    bbox_pred = u.box_cxcywh_to_xyxy(outputs_coord)
-                    prediction["bboxes"] = bbox_pred.tolist()
+                    if len(outputs_coord) == 0:
+                        prediction["bboxes"] = []
+                    else:
+                        bbox_pred = u.box_cxcywh_to_xyxy(outputs_coord)
+                        prediction["bboxes"] = bbox_pred.tolist()
                 else:
                     prediction["bboxes"] = []
+
                 if outputs_class is not None:
-                    result_class = torch.argmax(outputs_class, dim=1)
-                    prediction["classes"] = result_class.tolist()
+                    if len(outputs_class) == 0:
+                        prediction["classes"] = []
+                    else:
+                        result_class = torch.argmax(outputs_class, dim=1)
+                        prediction["classes"] = result_class.tolist()
                 else:
                     prediction["classes"] = []
                 if self._remove_padding:

{docling_ibm_models-1.2.0 → docling_ibm_models-1.3.0}/docling_ibm_models/tableformer/models/table04_rs/tablemodel04_rs.py

@@ -308,8 +308,12 @@ class TableModel04_rs(BaseModel, nn.Module):
 
         if len(outputs_coord1) > 0:
             outputs_coord1 = torch.stack(outputs_coord1)
+        else:
+            outputs_coord1 = torch.empty(0)
         if len(outputs_class1) > 0:
             outputs_class1 = torch.stack(outputs_class1)
+        else:
+            outputs_class1 = torch.empty(0)
 
         outputs_class = outputs_class1
         outputs_coord = outputs_coord1

{docling_ibm_models-1.2.0 → docling_ibm_models-1.3.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "docling-ibm-models"
-version = "1.2.0"  # DO NOT EDIT, updated automatically
+version = "1.3.0"  # DO NOT EDIT, updated automatically
 description = "This package contains the AI models used by the Docling PDF conversion package"
 authors = ["Nikos Livathinos <nli@zurich.ibm.com>", "Maxim Lysak <mly@zurich.ibm.com>", "Ahmed Nassar <ahn@zurich.ibm.com>", "Christoph Auer <cau@zurich.ibm.com>", "Michele Dolfi <dol@zurich.ibm.com>", "Peter Staar <taa@zurich.ibm.com>"]
 license = "MIT"
@@ -38,6 +38,7 @@ Pillow = "^10.0.0"
 tqdm = "^4.64.0"
 mean_average_precision = "^2021.4.26.0"
 opencv-python-headless = { version = "^4.9.0.80" }
+huggingface_hub = ">=0.23,<1"
 
 [tool.poetry.dev-dependencies]
 black = {extras = ["jupyter"], version = "^24.4.2"}