Uploaded subsets of Hagrid dataset with top X images per category

extract_subset_from_hagrid.py

@@ -0,0 +1,117 @@
+# %% Cell 1
+
+import os
+import json
+from typing import Any
+from zipfile import ZipFile
+
+def export_subset_of_hagrid(output_name: str, count_to_export: int = 1000):
+    image_subset_by_sign: dict[str, list[str]] = {}
+
+
+    with ZipFile('hagridv2_512.zip') as zf:
+        for file in zf.namelist():
+            if file.endswith('.jpg'):
+                sign_category = file.split("/")[1]
+                os.makedirs(f"{output_name}/{sign_category}", exist_ok=True)
+                if len(image_subset_by_sign.get(sign_category, [])) >= count_to_export:
+                    continue
+                file_name = file.split('/')[-1]
+                if sign_category not in image_subset_by_sign:
+                    image_subset_by_sign[sign_category] = []
+                image_subset_by_sign[sign_category].append(file_name[:-4])
+                export_file_name = f"{output_name}/{sign_category}/{file_name}"
+                with open(export_file_name, 'wb') as img_file:
+                    img_file.write(zf.read(file))
+
+    with open("exported_images.json", 'w') as file:
+        json.dump(image_subset_by_sign, file)
+
+# export_subset_of_hagrid('hagrid-10', count_to_export=10)
+
+# %% Cell 2
+
+import json
+
+def extract_connected_annotations(output_name: str):
+    with ZipFile("annotations.zip") as fa:
+
+        with open("exported_images.json", 'r') as file:
+            exported_examples = json.load(file)
+
+        annotations_for_exported_subset: dict[str, dict[str, Any]] = {}
+        minimal_annotations_for_exported_subset: dict[str, dict[str, Any]] = {}
+
+
+        for file in fa.namelist()[:]:
+            print(file)
+            if file.endswith('.json'):
+                category = file.split("/")[-1][:-5]
+                if category not in annotations_for_exported_subset:
+                    annotations_for_exported_subset[category] = {}
+                    minimal_annotations_for_exported_subset[category] = {}
+                if category not in exported_examples:
+                    continue
+
+                with fa.open(file) as json_file:
+                    labels = json.load(json_file)
+
+                    print(category)
+                    # print(exported_examples[category])
+                    print(set(labels.keys()).intersection(exported_examples[category]))
+
+                    for overlapping_images in set(labels.keys()).intersection(exported_examples[category]):
+                        annotations_for_exported_subset[category][overlapping_images] = labels[overlapping_images]
+                        minimal_annotations_for_exported_subset[category][overlapping_images] = {
+                            "labels": labels[overlapping_images]["labels"],
+                            "bboxes": labels[overlapping_images]["bboxes"],
+                            "meta": labels[overlapping_images]["meta"],
+                        }
+
+        with open(f"minimal-annotations.json", 'w') as f:
+            json.dump(minimal_annotations_for_exported_subset, f, indent=4)
+
+
+        with open(f"annotations.json", 'w') as f:
+            json.dump(annotations_for_exported_subset, f, indent=4)
+
+# extract_connected_annotations('hagrid-10')
+# %% Cell 3
+def run_pipeline(output_name: str, count_exported: int):
+    export_subset_of_hagrid(output_name, count_exported)
+    extract_connected_annotations(output_name)
+
+    annotation_files = [
+        "minimal-annotations.json",
+        "annotations.json",
+        "exported_images.json"
+    ]
+
+    output_zip_name = f"{output_name}_{count_exported}_images.zip"
+    with ZipFile(output_zip_name, 'w') as archive:
+        for root, dirs, files in os.walk(output_name):
+            for file in files:
+                file_path = os.path.join(root, file)
+                archive.write(file_path, arcname=os.path.relpath(file_path, start=output_name))
+
+        for file_name in annotation_files:
+            full_path = os.path.join(output_name, file_name)
+            if os.path.exists(full_path):
+                archive.write(full_path, arcname=file_name)
+
+    for root, dirs, files in os.walk(output_name, topdown=False):
+        for name in files:
+            os.remove(os.path.join(root, name))
+        for name in dirs:
+            os.rmdir(os.path.join(root, name))
+    os.rmdir(output_name)
+
+    for file_name in annotation_files:
+        print(file_name)
+        file_to_remove = os.path.join(output_name, file_name)
+        if os.path.exists(file_name):
+            os.remove(file_name)
+
+run_pipeline('hagrid-export', 100)
+run_pipeline('hagrid-export', 500)
+run_pipeline('hagrid-export', 1000)

hagrid-export_1000_images.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43505fb19cc383478d34c7242c3032beb957896bedfe9cfd4a354d848f8db92d
+size 4056532453

hagrid-export_100_images.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72ada2959c117d23e98c8c4da2c9009580554d12163a6db8f78de50312cbdc47
+size 402272913

hagrid-export_500_images.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9ce69320543d29f1e15c750b25a421019b2ded7798e6c850b3b203d3ddb0436
+size 2028554073

hagrid-export_5_images.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:615edd12fbe0f56cdaf8573a82b4c1b6ef4c5cf9a8a4c9a258bc4d5bc2948ffd
+size 20336296