fix: support empty tensors in cnn & transformer

aphp · Feb 15, 2024 · 1c0c095 · 1c0c095
1 parent 86f771d
commit 1c0c095
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 21 deletions.
diff --git a/edspdf/pipes/embeddings/huggingface_embedding.py b/edspdf/pipes/embeddings/huggingface_embedding.py
@@ -309,14 +309,15 @@ def collate(self, batch):
             "line_window_offsets_flat": line_window_offsets_flat,
         }
         if self.use_image:
-            collated["pixel_values"] = torch.stack(
+            collated["pixel_values"] = torch.as_tensor(
                 [
-                    torch.from_numpy(page_pixels)
+                    page_pixels
                     for sample_pages in batch["pixel_values"]
                     for page_pixels in sample_pages
                 ],
-                dim=0,
-            ).repeat_interleave(torch.as_tensor(windows_count_per_page), dim=0)
+            ).repeat_interleave(
+                torch.as_tensor(windows_count_per_page, dtype=torch.long), dim=0
+            )
         return collated
 
     def forward(self, batch):

diff --git a/tests/conftest.py b/tests/conftest.py
@@ -1,6 +1,4 @@
-import copy
 import os
-from functools import lru_cache
 from pathlib import Path
 
 import pytest
@@ -53,18 +51,6 @@ def error_pdf():
     return path.read_bytes()
 
 
-@lru_cache(maxsize=1)
-def make_pdfdoc(pdf):
-    from edspdf.pipes.extractors.pdfminer import PdfMinerExtractor
-
-    return PdfMinerExtractor(render_pages=True)(pdf)
-
-
-@fixture()
-def pdfdoc(pdf):
-    return copy.deepcopy(make_pdfdoc(pdf))
-
-
 @fixture(scope="session")
 def dummy_dataset(tmpdir_factory, pdf):
     tmp_path = tmpdir_factory.mktemp("datasets")

diff --git a/tests/pipes/embeddings/test_custom.py b/tests/pipes/embeddings/test_custom.py
@@ -3,9 +3,10 @@
 from edspdf.pipes.embeddings.embedding_combiner import EmbeddingCombiner
 from edspdf.pipes.embeddings.simple_text_embedding import SimpleTextEmbedding
 from edspdf.pipes.embeddings.sub_box_cnn_pooler import SubBoxCNNPooler
+from edspdf.pipes.extractors.pdfminer import PdfMinerExtractor
 
 
-def test_custom_embedding(pdfdoc, tmp_path):
+def test_custom_embedding(pdf, error_pdf, tmp_path):
     embedding = BoxTransformer(
         num_heads=4,
         dropout_p=0.1,
@@ -35,8 +36,14 @@ def test_custom_embedding(pdfdoc, tmp_path):
         ),
     )
     str(embedding)
+
+    extractor = PdfMinerExtractor(render_pages=True)
+    pdfdoc = extractor(pdf)
     pdfdoc.text_boxes[0].text = "Very long word of 150 letters : " + "x" * 150
     embedding.post_init([pdfdoc], set())
     embedding(pdfdoc)
     embedding.save_extra_data(tmp_path, set())
     embedding.load_extra_data(tmp_path, set())
+
+    # Test empty document
+    embedding(extractor(error_pdf))
diff --git a/tests/pipes/embeddings/test_huggingface.py b/tests/pipes/embeddings/test_huggingface.py
@@ -1,7 +1,8 @@
 from edspdf.pipes.embeddings.huggingface_embedding import HuggingfaceEmbedding
+from edspdf.pipes.extractors.pdfminer import PdfMinerExtractor
 
 
-def test_huggingface_embedding(pdfdoc):
+def test_huggingface_embedding(pdf, error_pdf):
     embedding = HuggingfaceEmbedding(
         pipeline=None,
         name="huggingface",
@@ -15,4 +16,7 @@ def test_huggingface_embedding(pdfdoc):
         "height": embedding.hf_model.config.input_size,
         "width": embedding.hf_model.config.input_size,
     }
-    embedding(pdfdoc)
+
+    extractor = PdfMinerExtractor(render_pages=True)
+    embedding(extractor(pdf))
+    embedding(extractor(error_pdf))