Merge pull request #223 from InnopolisUni/pose_estimation

pose estimation dataset and model

config/datasets/detection/coco_pose_recognition.yaml

@@ -0,0 +1,31 @@
+task:
+  - image-detection
+
+name: pose recognition dataset
+description: "COCO Pose dataset for human pose recognitoon. See more: https://cocodataset.org/#keypoints-2017"
+markup_info: There is a total of 2346 unique images, format compatible with ultralytics
+date_time: 22.05.2024
+
+_target_: innofw.core.integrations.ultralytics.datamodule.UltralyticsDataModuleAdapter
+
+train:
+  source: https://api.blackhole.ai.innopolis.university/public-datasets/coco_pose_recognition/train.zip
+  target: ./data/coco_pose/train/
+
+test:
+  source: https://api.blackhole.ai.innopolis.university/public-datasets/coco_pose_recognition/test.zip
+  target: ./data/coco_pose/test/
+
+infer:
+  source: https://api.blackhole.ai.innopolis.university/public-datasets/coco_pose_recognition/test.zip
+  target: ./data/coco_pose/test/
+
+num_workers: 8
+
+val_size: 0.2
+channels_num: 3
+image_size: 512
+num_classes: 1
+is_keypoint: True
+names:
+  - person

config/experiments/detection/KG_220524_coco.yaml

@@ -0,0 +1,21 @@
+# @package _global_
+defaults:
+- override /datasets: detection/coco_pose_recognition.yaml
+- override /models: detection/yolo8_pose.yaml
+
+# devices: 1
+epochs: 3
+accelerator: gpu
+batch_size: 32
+project: coco_pose_recognition
+random_seed: 0
+task: image-detection
+weights_freq: 1
+
+
+wandb:
+  enable: True
+  project: coco_pose_recognition
+  entity: "k-galliamov"
+  group: none
+  job_type: training

config/models/detection/yolo8_pose.yaml

@@ -0,0 +1,4 @@
+_target_: ultralytics.YOLO
+description: yolov8 pose by ultralytics
+model: yolov8n-pose.pt
+name: yolov8

innofw/core/integrations/ultralytics/datamodule.py

@@ -39,23 +39,23 @@ class UltralyticsDataModuleAdapter(BaseDataModule):
     framework = [Frameworks.ultralytics]
 
     def __init__(
-        self,
-        train: Optional[str],
-        # val: Optional[str],  # todo: add this
-        test: Optional[str],
-        infer: Optional[str],
-        num_workers: int,
-        image_size: int,
-        num_classes: int,
-        names: List[str],
-        batch_size: int = 4,
-        val_size: float = 0.2,
-        augmentations=None,
-        stage=False,
-        channels_num: int = 3,
-        random_state: int = 42,
-        *args,
-        **kwargs,
+            self,
+            train: Optional[str],
+            # val: Optional[str],  # todo: add this
+            test: Optional[str],
+            infer: Optional[str],
+            num_workers: int,
+            image_size: int,
+            num_classes: int,
+            names: List[str],
+            batch_size: int = 4,
+            val_size: float = 0.2,
+            augmentations=None,
+            stage=False,
+            channels_num: int = 3,
+            random_state: int = 42,
+            *args,
+            **kwargs,
     ):
         """
         Arguments:
@@ -73,7 +73,8 @@ def __init__(
         if self.train:
             self.train_source = Path(self.train)
             # # In this datamodule, the train source should be the folder train itself not the folder "train/images"
-            if str(self.train_source).endswith("images") or str(self.train_source).endswith("labels"):
+            if str(self.train_source).endswith("images") or str(self.train_source).endswith(
+                    "labels"):
                 self.train_source = Path(str(self.train_source)[:-7])
         if self.test:
             self.test_source = Path(self.test)
@@ -86,7 +87,8 @@ def __init__(
                 if not (type(self.infer) == str and self.infer.startswith("rts"))
                 else self.infer
             )
-            if str(self.infer_source).endswith("images") or str(self.infer_source).endswith("labels"):
+            if str(self.infer_source).endswith("images") or str(self.infer_source).endswith(
+                    "labels"):
                 self.infer_source = Path(str(self.infer_source)[:-7])
 
         self.batch_size = batch_size
@@ -97,6 +99,7 @@ def __init__(
         self.names = names
         self.random_state = random_state
         self.augmentations = augmentations
+        self.is_keypoint = kwargs.get("is_keypoint", False)
 
     def setup_train_test_val(self, **kwargs):
         """
@@ -146,7 +149,7 @@ def setup_train_test_val(self, **kwargs):
         img_files = list(train_img_path.iterdir())
         label_files = list(train_lbl_path.iterdir())
         assert (
-            len(label_files) == len(img_files) != 0
+                len(label_files) == len(img_files) != 0
         ), "number of images and labels should be the same"
 
         # sort the files so that the images and labels are in the same order
@@ -168,7 +171,7 @@ def setup_train_test_val(self, **kwargs):
 
         # Creating the images directory
         for files, folder_name in zip(
-            [train_img_files, val_img_files], ["train", "val"]
+                [train_img_files, val_img_files], ["train", "val"]
         ):
             # create a folder
             new_path = new_img_path / folder_name
@@ -180,7 +183,7 @@ def setup_train_test_val(self, **kwargs):
 
         # Creating the labels directory
         for files, folder_name in zip(
-            [train_label_files, val_label_files], ["train", "val"]
+                [train_label_files, val_label_files], ["train", "val"]
         ):
             # create a folder
             new_path = new_lbl_path / folder_name
@@ -201,14 +204,17 @@ def setup_train_test_val(self, **kwargs):
             file.write(f"val: {self.val_dataset}\n")
             file.write(f"test: {self.test_dataset}\n")
 
+            if self.is_keypoint:
+                file.write(
+                    "kpt_shape: [17, 3]\nflip_idx: [0, 2, 1, 4, 3, 6, 5, 8, 7, 10, 9, 12, 11, 14, 13, 16, 15]\n")
             file.write(f"nc: {self.num_classes}\n")
             file.write(f"names: {self.names}\n")
 
     def setup_infer(self):
         if (
-            type(self.infer_source) == str
-            and self.infer_source.startswith("rts")
-            or Path(self.infer_source).is_file()
+                type(self.infer_source) == str
+                and self.infer_source.startswith("rts")
+                or Path(self.infer_source).is_file()
         ):
             return
         # root_dir

pckg_util.py

@@ -30,7 +30,7 @@ def install_and_import(package, version="", params="", link="", packageimportnam
             [sys.executable, "-m", "pip", *installation_cmd_list]
         )
     finally:
-        if packageimportname is None:
+        if packageimportname is None or packageimportname == "":
             globals()[package] = importlib.import_module(package)
         else:
             globals()[packageimportname] = importlib.import_module(packageimportname)