diff --git a/README.md b/README.md
index 0c46f2b..3ce13ff 100644
--- a/README.md
+++ b/README.md
@@ -889,7 +889,18 @@ Get tasks and export as YOLO format files.
 
 ```python
 tasks = client.get_image_tasks(project="YOUR_PROJECT_SLUG")
-client.export_yolo(tasks)
+client.export_yolo(tasks, output_dir="YOUR_DIRECTROY")
+```
+
+Get tasks and export as YOLO format files with classes.txt
+You can use fixed classes.txt and arrange order of each annotaiton file's order
+
+```python
+project_slug = "YOUR_PROJECT_SLUG"
+tasks = client.get_image_tasks(project=project_slug)
+annotations = client.get_annotations(project=project_slug)
+classes = list(map(lambda annotation: annotation["value"], annotations))
+client.export_yolo(tasks=tasks, classes=classes, output_dir="YOUR_DIRECTROY")
 ```
 
 ### Pascal VOC
diff --git a/fastlabel/__init__.py b/fastlabel/__init__.py
index 17cbb57..36c6412 100644
--- a/fastlabel/__init__.py
+++ b/fastlabel/__init__.py
@@ -529,7 +529,8 @@ def create_video_classification_task(
         """
         endpoint = "tasks/video/classification"
         if not utils.is_video_supported_ext(file_path):
-            raise FastLabelInvalidException("Supported extensions are mp4.", 422)
+            raise FastLabelInvalidException(
+                "Supported extensions are mp4.", 422)
         file = utils.base64_encode(file_path)
         payload = {"project": project, "name": name, "file": file}
         if status:
@@ -651,14 +652,17 @@ def export_coco(self, tasks: list, output_dir: str = os.path.join("output", "coc
         with open(file_path, 'w') as f:
             json.dump(coco, f, indent=4, ensure_ascii=False)
 
-    def export_yolo(self, tasks: list, output_dir: str = os.path.join("output", "yolo")) -> None:
+    def export_yolo(self, tasks: list, classes: list = [], output_dir: str = os.path.join("output", "yolo")) -> None:
         """
         Convert tasks to YOLO format and export as files.
+        If you pass classes, classes.txt will be generated based on it .
+        If not , classes.txt will be generated based on passed tasks .(Annotations never used in your project will not be exported.)
 
         tasks is a list of tasks. (Required)
+        classes is a list of annotation values.  e.g. ['dog','bird'] (Optional)
         output_dir is output directory(default: output/yolo). (Optional)
         """
-        annos, categories = converters.to_yolo(tasks)
+        annos, categories = converters.to_yolo(tasks, classes)
         for anno in annos:
             file_name = anno["filename"]
             basename = utils.get_basename(file_name)
@@ -709,12 +713,12 @@ def export_labelme(self, tasks: list, output_dir: str = os.path.join("output", "
             with open(file_path, 'w') as f:
                 json.dump(labelme, f, indent=4, ensure_ascii=False)
 
-
     # Instance / Semantic Segmetation
+
     def export_instance_segmentation(self, tasks: list, output_dir: str = os.path.join("output", "instance_segmentation"), pallete: List[int] = const.COLOR_PALETTE) -> None:
         """
         Convert tasks to index color instance segmentation (PNG files).
-        Supports only bbox, polygon and segmentation annotation types. Hollowed points are not supported.
+        Supports only bbox, polygon and segmentation annotation types.
         Supports up to 57 instances in default colors palette. Check const.COLOR_PALETTE for more details.
 
         tasks is a list of tasks. (Required)
@@ -723,12 +727,13 @@ def export_instance_segmentation(self, tasks: list, output_dir: str = os.path.jo
         """
         tasks = converters.to_pixel_coordinates(tasks)
         for task in tasks:
-            self.__export_index_color_image(task=task, output_dir=output_dir, pallete=pallete, is_instance_segmentation=True)
-    
+            self.__export_index_color_image(
+                task=task, output_dir=output_dir, pallete=pallete, is_instance_segmentation=True)
+
     def export_semantic_segmentation(self, tasks: list, output_dir: str = os.path.join("output", "semantic_segmentation"), pallete: List[int] = const.COLOR_PALETTE) -> None:
         """
         Convert tasks to index color semantic segmentation (PNG files).
-        Supports only bbox, polygon and segmentation annotation types. Hollowed points are not supported.
+        Supports only bbox, polygon and segmentation annotation types.
         Check const.COLOR_PALETTE for color pallete.
 
         tasks is a list of tasks. (Required)
@@ -744,7 +749,8 @@ def export_semantic_segmentation(self, tasks: list, output_dir: str = os.path.jo
 
         tasks = converters.to_pixel_coordinates(tasks)
         for task in tasks:
-            self.__export_index_color_image(task=task, output_dir=output_dir, pallete=pallete, is_instance_segmentation=False, classes=classes)
+            self.__export_index_color_image(
+                task=task, output_dir=output_dir, pallete=pallete, is_instance_segmentation=False, classes=classes)
 
     def __export_index_color_image(self, task: list, output_dir: str, pallete: List[int], is_instance_segmentation: bool = True, classes: list = []) -> None:
         image = Image.new("RGB", (task["width"], task["height"]), 0)
@@ -753,28 +759,39 @@ def __export_index_color_image(self, task: list, output_dir: str, pallete: List[
 
         index = 1
         for annotation in task["annotations"]:
-            color = index if is_instance_segmentation else classes.index(annotation["value"]) + 1
+            color = index if is_instance_segmentation else classes.index(
+                annotation["value"]) + 1
             if annotation["type"] == AnnotationType.segmentation.value:
                 for region in annotation["points"]:
                     count = 0
                     for points in region:
-                        cv_draw_points = self.__get_cv_draw_points(points)
                         if count == 0:
-                            cv2.fillPoly(image, [cv_draw_points], color, lineType=cv2.LINE_8, shift=0)
+                            cv_draw_points = self.__get_cv_draw_points(points)
+                            cv2.fillPoly(
+                                image, [cv_draw_points], color, lineType=cv2.LINE_8, shift=0)
                         else:
-                            cv2.fillPoly(image, [cv_draw_points], 0, lineType=cv2.LINE_8, shift=0)
+                            # Reverse hollow points for opencv because this points are counter clockwise
+                            cv_draw_points = self.__get_cv_draw_points(
+                                utils.reverse_points(points))
+                            cv2.fillPoly(
+                                image, [cv_draw_points], 0, lineType=cv2.LINE_8, shift=0)
                         count += 1
             elif annotation["type"] == AnnotationType.polygon.value:
-                cv_draw_points = self.__get_cv_draw_points(annotation["points"])
-                cv2.fillPoly(image, [cv_draw_points], color, lineType=cv2.LINE_8, shift=0)
+                cv_draw_points = self.__get_cv_draw_points(
+                    annotation["points"])
+                cv2.fillPoly(image, [cv_draw_points], color,
+                             lineType=cv2.LINE_8, shift=0)
             elif annotation["type"] == AnnotationType.bbox.value:
-                cv_draw_points = self.__get_cv_draw_points(annotation["points"])
-                cv2.fillPoly(image, [cv_draw_points], color, lineType=cv2.LINE_8, shift=0)
+                cv_draw_points = self.__get_cv_draw_points(
+                    annotation["points"])
+                cv2.fillPoly(image, [cv_draw_points], color,
+                             lineType=cv2.LINE_8, shift=0)
             else:
                 continue
             index += 1
 
-        image_path = os.path.join(output_dir, utils.get_basename(task["name"]) + ".png")
+        image_path = os.path.join(
+            output_dir, utils.get_basename(task["name"]) + ".png")
         os.makedirs(os.path.dirname(image_path), exist_ok=True)
         image = Image.fromarray(image)
         image = image.convert('P')
@@ -823,7 +840,6 @@ def __get_cv_draw_points(self, points: List[int]) -> List[int]:
             cv_points.append((new_points[i * 2], new_points[i * 2 + 1]))
         return np.array(cv_points)
 
-
     # Annotation
 
     def find_annotation(self, annotation_id: str) -> dict:
diff --git a/fastlabel/converters.py b/fastlabel/converters.py
index ed56ae6..47b9409 100644
--- a/fastlabel/converters.py
+++ b/fastlabel/converters.py
@@ -4,6 +4,7 @@
 import copy
 import geojson
 import numpy as np
+import math
 from fastlabel.const import AnnotationType
 
 # COCO
@@ -144,10 +145,12 @@ def __calc_area(annotation_type: str, points: list) -> float:
 # YOLO
 
 
-def to_yolo(tasks: list) -> tuple:
-    coco = to_coco(tasks)
-    yolo = __coco2yolo(coco)
-    return yolo
+def to_yolo(tasks: list, classes: list) -> tuple:
+    if len(classes) == 0:
+        coco = to_coco(tasks)
+        return __coco2yolo(coco)
+    else:
+        return __to_yolo(tasks, classes)
 
 
 def __coco2yolo(coco: dict) -> tuple:
@@ -198,6 +201,69 @@ def __coco2yolo(coco: dict) -> tuple:
     return annos, categories
 
 
+def __to_yolo(tasks: list, classes: list) -> tuple:
+    annos = []
+    for task in tasks:
+        if task["height"] == 0 or task["width"] == 0:
+            continue
+        objs = []
+        data = [{"annotation": annotation, "task": task, "classes": classes}
+                for annotation in task["annotations"]]
+        with ThreadPoolExecutor(max_workers=8) as executor:
+            results = executor.map(__get_yolo_annotation, data)
+            for result in results:
+                if not result:
+                    continue
+                objs.append(" ".join(result))
+            anno = {
+                "filename": task["name"],
+                "object": objs
+            }
+            annos.append(anno)
+
+    categories = map(lambda val: {"name": val}, classes)
+
+    return annos, categories
+
+
+def __get_yolo_annotation(data: dict) -> dict:
+    annotation = data["annotation"]
+    points = annotation["points"]
+    annotation_type = annotation["type"]
+    value = annotation["value"]
+    classes = list(data["classes"])
+    task = data["task"]
+    if annotation_type != AnnotationType.bbox.value and annotation_type != AnnotationType.polygon.value:
+        return None
+    if not points or len(points) == 0:
+        return None
+    if annotation_type == AnnotationType.bbox.value and (int(points[0]) == int(points[2]) or int(points[1]) == int(points[3])):
+        return None
+    if not annotation["value"] in classes:
+        return None
+
+    dw = 1. / task["width"]
+    dh = 1. / task["height"]
+
+    bbox = __to_bbox(points)
+    xmin = bbox[0]
+    ymin = bbox[1]
+    xmax = bbox[0] + bbox[2]
+    ymax = bbox[1] + bbox[3]
+
+    x = (xmin + xmax) / 2
+    y = (ymin + ymax) / 2
+    w = xmax - xmin
+    h = ymax - ymin
+
+    x = str(_truncate(x * dw, 7))
+    y = str(_truncate(y * dh, 7))
+    w = str(_truncate(w * dw, 7))
+    h = str(_truncate(h * dh, 7))
+    category_index = str(classes.index(value))
+    return [category_index, x, y, w, h]
+
+
 def _truncate(n, decimals=0) -> float:
     multiplier = 10 ** decimals
     return int(n * multiplier) / multiplier
@@ -207,10 +273,75 @@ def _truncate(n, decimals=0) -> float:
 
 
 def to_pascalvoc(tasks: list) -> list:
-    coco = to_coco(tasks)
-    pascalvoc = __coco2pascalvoc(coco)
+    pascalvoc = []
+    for task in tasks:
+        if task["height"] == 0 or task["width"] == 0:
+            continue
+
+        pascal_objs = []
+        data = [{"annotation": annotation}
+                for annotation in task["annotations"]]
+        with ThreadPoolExecutor(max_workers=8) as executor:
+            results = executor.map(__get_pascalvoc_obj, data)
+
+        for result in results:
+            if not result:
+                continue
+            pascal_objs.append(result)
+
+        voc = {
+            "annotation": {
+                "filename": task["name"],
+                "size": {
+                    "width": task["width"],
+                    "height": task["height"],
+                    "depth": 3,
+                },
+                "segmented": 0,
+                "object": pascal_objs
+            }
+        }
+        pascalvoc.append(voc)
     return pascalvoc
 
+def __get_pascalvoc_obj(data: dict) -> dict:
+    annotation = data["annotation"]
+    points = annotation["points"]
+    annotation_type = annotation["type"]
+    if annotation_type != AnnotationType.bbox.value and annotation_type != AnnotationType.polygon.value:
+        return None
+    if not points or len(points) == 0:
+        return None
+    if annotation_type == AnnotationType.bbox.value and (int(points[0]) == int(points[2]) or int(points[1]) == int(points[3])):
+        return None
+    bbox = __to_bbox(points)
+    x = bbox[0]
+    y = bbox[1]
+    w = bbox[2]
+    h = bbox[3]
+
+    return {
+        "name": annotation["value"],
+        "pose": "Unspecified",
+                "truncated": __get_pascalvoc_tag_value(annotation, "truncated"),
+                "occluded": __get_pascalvoc_tag_value(annotation, "occluded"),
+                "difficult": __get_pascalvoc_tag_value(annotation, "difficult"),
+                "bndbox": {
+                    "xmin": math.floor(x),
+                    "ymin": math.floor(y),
+                    "xmax": math.floor(x + w),
+                    "ymax": math.floor(y + h),
+        },
+    }
+
+def __get_pascalvoc_tag_value(annotation: dict, target_tag_name: str) -> int:
+    attributes = annotation["attributes"]
+    if not attributes:
+        return 0
+    related_attr = next(
+        (attribute for attribute in attributes if attribute["type"] == "switch" and attribute["key"] == target_tag_name), None)
+    return int(related_attr["value"]) if related_attr else 0
+
 
 # labelme
 
@@ -388,58 +519,3 @@ def __get_pixel_coordinates(points: List[int or float]) -> List[int]:
                 new_points.append(int(prev_x + int(xdiff / mindiff * (i + 1))))
                 new_points.append(int(prev_y + int(ydiff / mindiff * (i + 1))))
     return new_points
-
-def __coco2pascalvoc(coco: dict) -> list:
-    pascalvoc = []
-
-    for image in coco["images"]:
-
-        # Get objects
-        objs = []
-        for annotation in coco["annotations"]:
-            if image["id"] != annotation["image_id"]:
-                continue
-            category = _get_category_by_id(
-                coco["categories"], annotation["category_id"])
-
-            x = annotation["bbox"][0]
-            y = annotation["bbox"][1]
-            w = annotation["bbox"][2]
-            h = annotation["bbox"][3]
-
-            obj = {
-                "name": category["name"],
-                "pose": "Unspecified",
-                "truncated": 0,
-                "difficult": 0,
-                "bndbox": {
-                        "xmin": x,
-                        "ymin": y,
-                        "xmax": x + w,
-                        "ymax": y + h,
-                },
-            }
-            objs.append(obj)
-
-        # get annotation
-        voc = {
-            "annotation": {
-                "filename": image["file_name"],
-                "size": {
-                    "width": image["width"],
-                    "height": image["height"],
-                    "depth": 3,
-                },
-                "segmented": 0,
-                "object": objs
-            }
-        }
-        pascalvoc.append(voc)
-
-    return pascalvoc
-
-
-def _get_category_by_id(categories: list, id_: str) -> str:
-    category = [
-        category for category in categories if category["id"] == id_][0]
-    return category
diff --git a/fastlabel/utils.py b/fastlabel/utils.py
index 7984222..58d9254 100644
--- a/fastlabel/utils.py
+++ b/fastlabel/utils.py
@@ -21,3 +21,18 @@ def get_basename(file_path: str) -> str:
           path/to/file.jpg -> path/to/file
     """
     return os.path.splitext(file_path)[0]
+
+
+def reverse_points(points: list[int]) -> list[int]:
+    """
+    e.g.)
+    [4, 5, 4, 9, 8, 9, 8, 5, 4, 5] => [4, 5, 8, 5, 8, 9, 4, 9, 4, 5]
+    """
+    reversed_points = []
+    for index, _ in enumerate(points):
+        if index % 2 == 0:
+            reversed_points.insert(
+                0, points[index + 1])
+            reversed_points.insert(
+                0, points[index])
+    return reversed_points
diff --git a/setup.py b/setup.py
index 1beaa53..f22dd75 100644
--- a/setup.py
+++ b/setup.py
@@ -8,7 +8,7 @@
 
 setuptools.setup(
     name="fastlabel",
-    version="0.11.1",
+    version="0.11.4",
     author="eisuke-ueta",
     author_email="eisuke.ueta@fastlabel.ai",
     description="The official Python SDK for FastLabel API, the Data Platform for AI",