paddlex 2.0.0rc4__py3-none-any.whl → 3.0.0__py3-none-any.whl

paddlex/modules/video_classification/dataset_checker/dataset_src/check_dataset.py

@@ -0,0 +1,120 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import os.path as osp
+import random
+from collections import defaultdict
+
+from .....utils.errors import CheckFailedError, DatasetFileNotFoundError
+
+
+def check(dataset_dir, output, sample_num=10):
+    """check dataset"""
+    dataset_dir = osp.abspath(dataset_dir)
+    # Custom dataset
+    if not osp.exists(dataset_dir) or not osp.isdir(dataset_dir):
+        raise DatasetFileNotFoundError(file_path=dataset_dir)
+
+    tags = ["train", "val"]
+    delim = " "
+    valid_num_parts = 2
+
+    sample_cnts = dict()
+    label_map_dict = dict()
+    sample_paths = defaultdict(list)
+    labels = []
+
+    label_file = osp.join(dataset_dir, "label.txt")
+    if not osp.exists(label_file):
+        raise DatasetFileNotFoundError(
+            file_path=label_file,
+            solution=f"Ensure that `label.txt` exist in {dataset_dir}",
+        )
+
+    with open(label_file, "r", encoding="utf-8") as f:
+        all_lines = f.readlines()
+        for line in all_lines:
+            substr = line.strip("\n").split(" ", 1)
+            try:
+                label_idx = int(substr[0])
+                labels.append(label_idx)
+                label_map_dict[label_idx] = str(substr[1])
+            except:
+                raise CheckFailedError(
+                    f"Ensure that the first number in each line in {label_file} should be int."
+                )
+    if min(labels) != 0:
+        raise CheckFailedError(
+            f"Ensure that the index starts from 0 in `{label_file}`."
+        )
+
+    for tag in tags:
+        file_list = osp.join(dataset_dir, f"{tag}.txt")
+        if not osp.exists(file_list):
+            if tag in ("train", "val"):
+                # train and val file lists must exist
+                raise DatasetFileNotFoundError(
+                    file_path=file_list,
+                    solution=f"Ensure that both `train.txt` and `val.txt` exist in {dataset_dir}",
+                )
+            else:
+                # tag == 'test'
+                continue
+        else:
+            with open(file_list, "r", encoding="utf-8") as f:
+                all_lines = f.readlines()
+                random.seed(123)
+                random.shuffle(all_lines)
+                sample_cnts[tag] = len(all_lines)
+                for line in all_lines:
+                    substr = line.strip("\n").split(delim)
+                    if len(substr) != valid_num_parts:
+                        raise CheckFailedError(
+                            f"The number of delimiter-separated items in each row in {file_list} \
+                                    should be {valid_num_parts} (current delimiter is '{delim}')."
+                        )
+                    file_name = substr[0]
+                    label = substr[1]
+
+                    video_path = osp.join(dataset_dir, file_name)
+
+                    if not osp.exists(video_path):
+                        raise DatasetFileNotFoundError(file_path=video_path)
+
+                    if len(sample_paths[tag]) < sample_num:
+                        sample_path = osp.join(
+                            "check_dataset", os.path.relpath(video_path, output)
+                        )
+                        sample_paths[tag].append(sample_path)
+
+                    try:
+                        label = int(label)
+                    except (ValueError, TypeError) as e:
+                        raise CheckFailedError(
+                            f"Ensure that the second number in each line in {label_file} should be int."
+                        ) from e
+
+    num_classes = max(labels) + 1
+
+    attrs = {}
+    attrs["label_file"] = osp.relpath(label_file, output)
+    attrs["num_classes"] = num_classes
+    attrs["train_samples"] = sample_cnts["train"]
+    attrs["train_sample_paths"] = sample_paths["train"]
+
+    attrs["val_samples"] = sample_cnts["val"]
+    attrs["val_sample_paths"] = sample_paths["val"]
+
+    return attrs

paddlex/modules/video_classification/dataset_checker/dataset_src/split_dataset.py

@@ -0,0 +1,82 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import os
+from random import shuffle
+
+from .....utils.file_interface import custom_open
+
+
+def split_dataset(root_dir, train_rate, val_rate):
+    """
+    Split the image dataset into training, validation, and test sets according to the given ratios,
+    and generate corresponding .txt files.
+
+    Args:
+        root_dir (str): Path to the root directory of the dataset.
+        train_rate (int): Percentage of the dataset to be used as the training set.
+        val_rate (int): Percentage of the dataset to be used as the validation set.
+
+    Returns:
+        str: Information about the dataset split results.
+    """
+    sum_rate = train_rate + val_rate
+    assert (
+        sum_rate == 100
+    ), f"The sum of train_rate({train_rate}), val_rate({val_rate}) should equal 100!"
+    assert (
+        train_rate > 0 and val_rate > 0
+    ), f"The train_rate({train_rate}) and val_rate({val_rate}) should be greater than 0!"
+    tags = ["train", "val"]
+    valid_path = False
+    video_files = []
+    for tag in tags:
+        split_image_list = os.path.abspath(os.path.join(root_dir, f"{tag}.txt"))
+        rename_image_list = os.path.abspath(os.path.join(root_dir, f"{tag}.txt.bak"))
+        if os.path.exists(split_image_list):
+            with custom_open(split_image_list, "r") as f:
+                lines = f.readlines()
+            video_files = video_files + lines
+            valid_path = True
+            if not os.path.exists(rename_image_list):
+                os.rename(split_image_list, rename_image_list)
+
+    assert (
+        valid_path
+    ), f"The files to be divided{tags[0]}.txt, {tags[1]}.txt, do not exist in the dataset directory."
+
+    shuffle(video_files)
+    start = 0
+    video_num = len(video_files)
+    rate_list = [train_rate, val_rate]
+    for i, tag in enumerate(tags):
+
+        rate = rate_list[i]
+        if rate == 0:
+            continue
+
+        end = start + round(video_num * rate / 100)
+        if sum(rate_list[i + 1 :]) == 0:
+            end = video_num
+
+        txt_file = os.path.abspath(os.path.join(root_dir, tag + ".txt"))
+        with custom_open(txt_file, "w") as f:
+            m = 0
+            for id in range(start, end):
+                m += 1
+                f.write(video_files[id])
+        start = end
+
+    return root_dir

paddlex/modules/video_classification/evaluator.py

@@ -0,0 +1,44 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ..base import BaseEvaluator
+from .model_list import MODELS
+
+
+class VideoClsEvaluator(BaseEvaluator):
+    """Image Classification Model Evaluator"""
+
+    entities = MODELS
+
+    def update_config(self):
+        """update evaluation config"""
+        if self.eval_config.log_interval:
+            self.pdx_config.update_log_interval(self.eval_config.log_interval)
+        self.pdx_config.update_dataset(
+            self.global_config.dataset_dir, "VideoClsDataset"
+        )
+        if self.eval_config.batch_size is not None:
+            self.pdx_config.update_batch_size(self.eval_config.batch_size, mode="eval")
+        self.pdx_config.update_pretrained_weights(self.eval_config.weight_path)
+
+    def get_eval_kwargs(self) -> dict:
+        """get key-value arguments of model evaluation function
+
+        Returns:
+            dict: the arguments of evaluation function.
+        """
+        return {
+            "weight_path": self.eval_config.weight_path,
+            "device": self.get_device(using_device_number=1),
+        }

paddlex/modules/video_classification/exportor.py

@@ -0,0 +1,22 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ..base import BaseExportor
+from .model_list import MODELS
+
+
+class VideoClsExportor(BaseExportor):
+    """Image Classification Model Exportor"""
+
+    entities = MODELS

paddlex/modules/video_classification/model_list.py

@@ -0,0 +1,19 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+MODELS = [
+    "PP-TSM-R50_8frames_uniform",
+    "PP-TSMv2-LCNetV2_8frames_uniform",
+    "PP-TSMv2-LCNetV2_16frames_uniform",
+]

paddlex/modules/video_classification/trainer.py

@@ -0,0 +1,88 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import shutil
+from pathlib import Path
+
+from ..base import BaseTrainer
+from .model_list import MODELS
+
+
+class VideoClsTrainer(BaseTrainer):
+    """Image Classification Model Trainer"""
+
+    entities = MODELS
+
+    def dump_label_dict(self, src_label_dict_path: str):
+        """dump label dict config
+
+        Args:
+            src_label_dict_path (str): path to label dict file to be saved.
+        """
+        dst_label_dict_path = Path(self.global_config.output).joinpath("label_dict.txt")
+        shutil.copyfile(src_label_dict_path, dst_label_dict_path)
+
+    def update_config(self):
+        """update training config"""
+        if self.train_config.log_interval:
+            self.pdx_config.update_log_interval(self.train_config.log_interval)
+        if self.train_config.eval_interval:
+            self.pdx_config.update_eval_interval(self.train_config.eval_interval)
+        if self.train_config.save_interval:
+            self.pdx_config.update_save_interval(self.train_config.save_interval)
+
+        self.pdx_config.update_dataset(
+            self.global_config.dataset_dir, "VideoClsDataset"
+        )
+        if self.train_config.num_classes is not None:
+            self.pdx_config.update_num_classes(self.train_config.num_classes)
+        if self.train_config.pretrain_weight_path != "":
+            self.pdx_config.update_pretrained_weights(
+                self.train_config.pretrain_weight_path
+            )
+
+        label_dict_path = Path(self.global_config.dataset_dir).joinpath("label.txt")
+        if label_dict_path.exists():
+            self.dump_label_dict(label_dict_path)
+        if self.train_config.batch_size is not None:
+            self.pdx_config.update_batch_size(
+                self.train_config.batch_size, mode="train"
+            )
+        if self.eval_config.batch_size is not None:
+            self.pdx_config.update_batch_size(self.eval_config.batch_size, mode="eval")
+        if self.train_config.learning_rate is not None:
+            self.pdx_config.update_learning_rate(self.train_config.learning_rate)
+        if self.train_config.epochs_iters is not None:
+            self.pdx_config._update_epochs(self.train_config.epochs_iters)
+        if self.train_config.warmup_steps is not None:
+            self.pdx_config.update_warmup_epochs(self.train_config.warmup_steps)
+        if self.global_config.output is not None:
+            self.pdx_config._update_output_dir(self.global_config.output)
+
+    def get_train_kwargs(self) -> dict:
+        """get key-value arguments of model training function
+
+        Returns:
+            dict: the arguments of training function.
+        """
+        train_args = {"device": self.get_device()}
+        if (
+            self.train_config.resume_path is not None
+            and self.train_config.resume_path != ""
+        ):
+            train_args["resume_path"] = self.train_config.resume_path
+        train_args["dy2st"] = self.train_config.get("dy2st", False)
+        # amp support 'O1', 'O2', 'OFF'
+        train_args["amp"] = self.train_config.get("amp", "OFF")
+        return train_args

paddlex/modules/video_detection/__init__.py

@@ -0,0 +1,18 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .dataset_checker import VideoDetDatasetChecker
+from .evaluator import VideoDetEvaluator
+from .exportor import VideoDetExportor
+from .trainer import VideoDetTrainer

paddlex/modules/video_detection/dataset_checker/__init__.py

@@ -0,0 +1,86 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from pathlib import Path
+
+from ...base import BaseDatasetChecker
+from ..model_list import MODELS
+from .dataset_src import check, deep_analyse
+
+
+class VideoDetDatasetChecker(BaseDatasetChecker):
+    """Dataset Checker for Image Classification Model"""
+
+    entities = MODELS
+    sample_num = 10
+
+    def convert_dataset(self, src_dataset_dir: str) -> str:
+        """convert the dataset from other type to specified type
+
+        Args:
+            src_dataset_dir (str): the root directory of dataset.
+
+        Returns:
+            str: the root directory of converted dataset.
+        """
+        return src_dataset_dir
+
+    def split_dataset(self, src_dataset_dir: str) -> str:
+        """repartition the train and validation dataset
+
+        Args:
+            src_dataset_dir (str): the root directory of dataset.
+
+        Returns:
+            str: the root directory of splited dataset.
+        """
+        return src_dataset_dir
+
+    def check_dataset(self, dataset_dir: str, sample_num: int = sample_num) -> dict:
+        """check if the dataset meets the specifications and get dataset summary
+
+        Args:
+            dataset_dir (str): the root directory of dataset.
+            sample_num (int): the number to be sampled.
+        Returns:
+            dict: dataset summary.
+        """
+        return check(dataset_dir, self.output)
+
+    def analyse(self, dataset_dir: str) -> dict:
+        """deep analyse dataset
+
+        Args:
+            dataset_dir (str): the root directory of dataset.
+
+        Returns:
+            dict: the deep analysis results.
+        """
+        return deep_analyse(dataset_dir, self.output)
+
+    def get_show_type(self) -> str:
+        """get the show type of dataset
+
+        Returns:
+            str: show type
+        """
+        return "video"
+
+    def get_dataset_type(self) -> str:
+        """return the dataset type
+
+        Returns:
+            str: dataset type
+        """
+        return "VideoDetDataset"

paddlex/modules/video_detection/dataset_checker/dataset_src/__init__.py

@@ -0,0 +1,17 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from .analyse_dataset import deep_analyse
+from .check_dataset import check

paddlex/modules/video_detection/dataset_checker/dataset_src/analyse_dataset.py

@@ -0,0 +1,100 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import platform
+from collections import defaultdict
+
+import numpy as np
+
+from .....utils.deps import function_requires_deps, is_dep_available
+from .....utils.file_interface import custom_open
+from .....utils.fonts import PINGFANG_FONT_FILE_PATH
+
+if is_dep_available("matplotlib"):
+    import matplotlib.pyplot as plt
+    from matplotlib import font_manager
+
+
+@function_requires_deps("matplotlib")
+def deep_analyse(dataset_path, output):
+    """class analysis for dataset"""
+    tags = ["train", "val"]
+    labels_cnt = defaultdict(str)
+    label_path = os.path.join(dataset_path, "label_map.txt")
+    with custom_open(label_path, "r") as f:
+        lines = f.readlines()
+    for line in lines:
+        line = line.strip().split()
+        labels_cnt[line[1]] = " ".join(line[0])
+    for tag in tags:
+        anno_path = os.path.join(dataset_path, f"{tag}.txt")
+        classes_num = defaultdict(int)
+        for i in range(len(labels_cnt)):
+            classes_num[labels_cnt[str(i)]] = 0
+        with custom_open(anno_path, "r") as f:
+            lines = f.readlines()
+        for line in lines:
+            line = line.strip().split()
+            label_file_path = os.path.join(dataset_path, line[0])
+            with custom_open(label_file_path, "r") as f:
+                label_lines = f.readlines()
+                for label_line in label_lines:
+                    label_info = label_line.strip().split(" ")
+                    classes_num[labels_cnt[label_info[0]]] += 1
+
+        if tag == "train":
+            cnts_train = [cat_ids for cat_name, cat_ids in classes_num.items()]
+        elif tag == "val":
+            cnts_val = [cat_ids for cat_name, cat_ids in classes_num.items()]
+
+    classes = [cat_name for cat_name, cat_ids in classes_num.items()]
+    sorted_id = sorted(
+        range(len(cnts_train)), key=lambda k: cnts_train[k], reverse=True
+    )
+    cnts_train_sorted = [cnts_train[index] for index in sorted_id]
+    cnts_val_sorted = [cnts_val[index] for index in sorted_id]
+    classes_sorted = [classes[index] for index in sorted_id]
+    x = np.arange(len(classes))
+    width = 0.5
+
+    # bar
+    os_system = platform.system().lower()
+    if os_system == "windows":
+        plt.rcParams["font.sans-serif"] = "FangSong"
+    else:
+        font = font_manager.FontProperties(fname=PINGFANG_FONT_FILE_PATH, size=10)
+    fig, ax = plt.subplots(figsize=(max(8, int(len(classes) / 5)), 5), dpi=300)
+    ax.bar(x, cnts_train_sorted, width=0.5, label="train")
+    ax.bar(x + width, cnts_val_sorted, width=0.5, label="val")
+    plt.xticks(
+        x + width / 2,
+        classes_sorted,
+        rotation=90,
+        fontproperties=None if os_system == "windows" else font,
+    )
+    ax.set_xlabel(
+        "类别名称", fontproperties=None if os_system == "windows" else font, fontsize=12
+    )
+    ax.set_ylabel(
+        "样本框数量",
+        fontproperties=None if os_system == "windows" else font,
+        fontsize=12,
+    )
+    plt.legend(loc=1)
+    fig.tight_layout()
+    file_path = os.path.join(output, "histogram.png")
+    fig.savefig(file_path, dpi=300)
+
+    return {"histogram": os.path.join("check_dataset", "histogram.png")}