adding specific format parsing: CVAT

Signed-off-by: Peter Staar <[email protected]>

docling_eval/benchmarks/formats/cvat.py

@@ -0,0 +1,67 @@
+from pathlib import Path
+
+def from_cvat_to_docling_document(annotation_filenames:List[Path], page_images:Path):
+    pass
+
+
+def from_docling_document_to_cvat(docs: List[DoclingDocument],
+                                  output_dir:Path,
+                                  labels: Set[DocItemLabel] = DEFAULT_EXPORT_LABELS):    
+
+    output_page_dir = output_dir / "page_images"
+
+    for _ in [output_dir, output_page_dir]:
+        os.makedirs(_, exist_ok=True)
+
+    img_id = 0
+    for doc_id,doc in enumerate(docs):
+
+        page_images = []
+        page_fnames = []
+        for j,page in enumerate(doc.pages):
+            page_image = page.pil_image
+            page_image.save(str(output_page_dir / f"doc_{i:06}_{j:06}.png"))
+
+            page_images.append(page_image)
+            page_fnames.append(f"doc_{i:06}_{j:06}.png")
+
+        page_bboxes = {i: [] for i,fname in enumerate(page_fnames)}
+        for item, level in doc.iterate_items():
+            if isinstance(item, DocItem) and item.label in labels:
+                for prov in item.prov:
+                    page_no = prov.page_no
+
+                    page_w = doc.pages[prov.page_no].size.width
+                    page_i = doc.pages[prov.page_no].size.height
+
+                    img_w = page_images[page_no-1].width
+                    img_h = page_images[page_no-1].height
+
+                    page_bbox = prov.bbox 
+                    origin = bbox.coord_origin
+
+                    img_bbox = [
+                        page_bbox.l/page_w*img_w,
+                        page_bbox.b/page_h*img_h,
+                        page_bbox.r/page_w*img_w,
+                        page_bbox.t/page_h*img_h,
+                    ]
+
+                    page_bboxes[page_no-1].append({
+                        "label": item.label,
+                        "bbox": img_bbox,
+                    })
+
+
+
+        results=[]
+        for page_no, page_file in enumerate(page_fnames):
+            img_w = page_images[page_no].width
+            img_h = page_images[page_no].height
+
+            results.append(f'<image id="{img_id}" name="{page_file}" width="{img_w}" height="{img_h}">')
+
+            for bbox_id, page_bbox in enumerate(page_bboxes):
+                results.append(f'<bbox label="{label}" source="manual" occluded="0" xtl="{}" ytl="{}" xbr="{}" ybr="{}" z_order="{bbox_id}">')
+
+            results.append("</image>")