【Hackathon + No.150】PaddleRS集成PaddleDetection的旋转框检测能力

docs/apis/data_cn.md

@@ -57,13 +57,14 @@
 |-------|----|--------|-----|
 |`data_dir`|`str`|数据集存放目录。||
 |`image_dir`|`str`|输入图像存放目录。||
-|`ann_path`|`str`|[COCO格式](https://cocodataset.org/#home)标注文件路径。||
+|`anno_path`|`str`|[COCO格式](https://cocodataset.org/#home)标注文件路径。||
 |`transforms`|`paddlers.transforms.Compose`|对输入数据应用的数据变换算子。||
 |`label_list`|`str` \| `None`|label list文件。label list是一个文本文件，其中每一行包含一个类别的名称。|`None`|
 |`num_workers`|`int` \| `str`|加载数据时使用的辅助进程数。若设置为`'auto'`，则按照如下规则确定使用进程数：当CPU核心数大于16时，使用8个数据读取辅助进程；否则，使用CPU核心数一半数量的辅助进程。|`'auto'`|
 |`shuffle`|`bool`|是否随机打乱数据集中的样本。|`False`|
 |`allow_empty`|`bool`|是否向数据集中添加负样本。|`False`|
 |`empty_ratio`|`float`|负样本占比，仅当`allow_empty`为`True`时生效。若`empty_ratio`为负值或大于等于1，则保留所有生成的负样本。|`1.0`|
+|`batch_transforms`|`paddlers.transforms.BatchCompose`|对输入数据应用的批数据变换算子。||
 
 ### VOC格式目标检测数据集`VOCDetDataset`
 
@@ -81,6 +82,7 @@
 |`shuffle`|`bool`|是否随机打乱数据集中的样本。|`False`|
 |`allow_empty`|`bool`|是否向数据集中添加负样本。|`False`|
 |`empty_ratio`|`float`|负样本占比，仅当`allow_empty`为`True`时生效。若`empty_ratio`为负值或大于等于1，则保留所有生成的负样本。|`1.0`|
+|`batch_transforms`|`paddlers.transforms.BatchCompose`|对输入数据应用的批数据变换算子。||
 
 `VOCDetDataset`对file list的要求如下：
 

docs/apis/data_en.md

@@ -57,13 +57,14 @@ The initialization parameter list is as follows:
 |-------|----|--------|-----|
 |`data_dir`|`str`|Directory that stores the dataset.||
 |`image_dir`|`str`|Directory of input images.||
-|`ann_path`|`str`|[COCO Format](https://cocodataset.org/#home)label file path.||
+|`anno_path`|`str`|[COCO Format](https://cocodataset.org/#home)label file path.||
 |`transforms`|`paddlers.transforms.Compose`|Data transformation operators applied to input data.||
 |`label_list`|`str` \| `None`|Label list path. Label list is a text file, in which each line contains the name of class.|`None`|
 |`num_workers`|`int` \| `str`|Number of auxiliary processes used when loading data. If it is set to `'auto'`, use the following rules to determine the number of processes to use: When the number of CPU cores is greater than 16, 8 data read auxiliary processes are used; otherwise, the number of auxiliary processes is set to half the counts of CPU cores.|`'auto'`|
 |`shuffle`|`bool`|Whether to randomly shuffle the samples in the dataset.|`False`|
 |`allow_empty`|`bool`|Whether to add negative samples to the dataset.|`False`|
 |`empty_ratio`|`float`|Negative sample ratio. Take effect only if `allow_empty` is `True`. If `empty_ratio` is negative or greater than or equal to 1, all negative samples generated are retained.|`1.0`|
+|`batch_transforms`|`paddlers.transforms.BatchCompose`|Data batch transformation operators applied to input data.||
 
 ### VOC Format Object Detection Dataset `VOCDetDataset`
 
@@ -81,6 +82,7 @@ The initialization parameter list is as follows:
 |`shuffle`|`bool`|Whether to randomly shuffle the samples in the dataset.|`False`|
 |`allow_empty`|`bool`|Whether to add negative samples to the dataset.|`False`|
 |`empty_ratio`|`float`|Negative sample ratio. Takes effect only if `allow_empty` is `True`. If `empty_ratio` is negative or greater than or equal to `1`, all negative samples generated will be retained.|`1.0`|
+|`batch_transforms`|`paddlers.transforms.BatchCompose`|Data batch transformation operators applied to input data.||
 
 The requirements of `VOCDetDataset` for the file list are as follows:
 

docs/apis/train_cn.md

@@ -166,6 +166,7 @@ def train(self,
           warmup_start_lr=0.0,
           lr_decay_epochs=(216, 243),
           lr_decay_gamma=0.1,
+          cosine_decay_num_epochs=1000,
           metric=None,
           use_ema=False,
           early_stop=False,
@@ -196,7 +197,8 @@ def train(self,
 |`warmup_start_lr`|`int`|默认优化器warm-up阶段使用的初始学习率。|`0`|
 |`lr_decay_epochs`|`list` \| `tuple`|默认优化器学习率衰减的milestones，以epoch计。即，在第几个epoch执行学习率的衰减。|`(216, 243)`|
 |`lr_decay_gamma`|`float`|学习率衰减系数，适用于默认优化器。|`0.1`|
-|`metric`|`str` \| `None`|评价指标，可以为`'VOC'`、`COCO`或`None`。若为`None`，则根据数据集格式自动确定使用的评价指标。|`None`|
+|`cosine_decay_num_epochs`|`int`|使用余弦退火学习率调度器时计算退火周期的参数。|`1000`|
+|`metric`|`str` \| `None`|评价指标，可以为`'VOC'`、`'COCO'`、`'RBOX'`或`None`。若为`None`，则根据数据集格式自动确定使用的评价指标。|`None`|
 |`use_ema`|`bool`|是否启用[指数滑动平均策略](https://github.com/PaddlePaddle/PaddleRS/blob/develop/paddlers/models/ppdet/optimizer.py)更新模型权重参数。|`False`|
 |`early_stop`|`bool`|训练过程是否启用早停策略。|`False`|
 |`early_stop_patience`|`int`|启用早停策略时的`patience`参数（参见[`EarlyStop`](https://github.com/PaddlePaddle/PaddleRS/blob/develop/paddlers/utils/utils.py)）。|`5`|

docs/apis/train_en.md

@@ -166,6 +166,7 @@ def train(self,
           warmup_start_lr=0.0,
           lr_decay_epochs=(216, 243),
           lr_decay_gamma=0.1,
+          cosine_decay_num_epochs=1000,
           metric=None,
           use_ema=False,
           early_stop=False,
@@ -196,7 +197,8 @@ The meaning of each parameter is as follows:
 |`warmup_start_lr`|`int`|Default initial learning rate used in the warm-up phase of the optimizer.|`0`|
 |`lr_decay_epochs`|`list` \| `tuple`|Milestones of learning rate decline of the default optimizer, in terms of epochs. That is, which epoch the decay of the learning rate occurs.|`(216, 243)`|
 |`lr_decay_gamma`|`float`|Learning rate attenuation coefficient, for default optimizer.|`0.1`|
-|`metric`|`str` \| `None`|Evaluation metrics, which can be `'VOC'`, `COCO`, or `None`. If `None`, the evaluation metrics will be automatically determined according to the format of the dataset.|`None`|
+|`cosine_decay_num_epochs`|`int`|Parameter to determine the annealing cycle when a cosine annealing learning rate scheduler is used.|`1000`|
+|`metric`|`str` \| `None`|Evaluation metrics, which can be `'VOC'`, `'COCO'`, `'RBOX'`, or `None`. If `None`, the evaluation metrics will be automatically determined according to the format of the dataset.|`None`|
 |`use_ema`|`bool`|Whether to enable [exponential moving average strategy](https://github.com/PaddlePaddle/PaddleRS/blob/develop/paddlers/models/ppdet/optimizer.py) to update model weights.|`False`|
 |`early_stop`|`bool`|Whether to enable the early stopping policy during training.|`False`|
 |`early_stop_patience`|`int`|`patience` parameter when the early stopping policy is enabled. Please refer to [`EarlyStop`](https://github.com/PaddlePaddle/PaddleRS/blob/develop/paddlers/utils/utils.py) for more details.|`5`|

docs/intro/data_prep_cn.md

@@ -9,5 +9,6 @@
 | 变化检测 | LEVIR-CD | https://justchenhao.github.io/LEVIR/ | [prepare_levircd.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_levircd.py) |
 | 变化检测 | Season-varying | https://paperswithcode.com/dataset/cdd-dataset-season-varying | [prepare_svcd.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_svcd.py) |
 | 场景分类 | UC Merced | http://weegee.vision.ucmerced.edu/datasets/landuse.html | [prepare_ucmerced.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_ucmerced.py) |
-| 目标检测 | RSOD | https://github.com/RSIA-LIESMARS-WHU/RSOD-Dataset- | [prepare_rsod](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_rsod.py) |
-| 图像分割 | iSAID | https://captain-whu.github.io/iSAID/ | [prepare_isaid](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_isaid.py) |
+| 目标检测 | DOTA | https://captain-whu.github.io/DOTA/ | [prepare_dota.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_dota.py) |
+| 目标检测 | RSOD | https://github.com/RSIA-LIESMARS-WHU/RSOD-Dataset- | [prepare_rsod.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_rsod.py) |
+| 图像分割 | iSAID | https://captain-whu.github.io/iSAID/ | [prepare_isaid.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_isaid.py) |

docs/intro/data_prep_en.md

@@ -9,5 +9,6 @@
 | Change Detection | LEVIR-CD | https://justchenhao.github.io/LEVIR/ | [prepare_levircd.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_levircd.py) |
 | Change Detection | Season-varying | https://paperswithcode.com/dataset/cdd-dataset-season-varying | [prepare_svcd.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_svcd.py) |
 | Scene Classification | UC Merced | http://weegee.vision.ucmerced.edu/datasets/landuse.html | [prepare_ucmerced.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_ucmerced.py) |
-| Object Detection | RSOD | https://github.com/RSIA-LIESMARS-WHU/RSOD-Dataset- | [prepare_rsod](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_rsod.py) |
-| Image Segmentation | iSAID | https://captain-whu.github.io/iSAID/ | [prepare_isaid](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_isaid.py) |
+| Object Detection | DOTA | https://captain-whu.github.io/DOTA/ | [prepare_dota.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_dota.py) |
+| Object Detection | RSOD | https://github.com/RSIA-LIESMARS-WHU/RSOD-Dataset- | [prepare_rsod.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_rsod.py) |
+| Image Segmentation | iSAID | https://captain-whu.github.io/iSAID/ | [prepare_isaid.py](https://github.com/PaddlePaddle/PaddleRS/blob/develop/tools/prepare_dataset/prepare_isaid.py) |

docs/intro/model_cons_params_cn.md

@@ -449,6 +449,7 @@
 
 | 参数名 | 描述                            | 默认值 |
 | --- |-------------------------------| --- |
+| `rotate (bool)` | 表示是否执行旋转目标检测 | `False` |
 | `num_classes (int)` | 目标类别数量                        | `80` |
 | `backbone (str)` | 骨干网络名称                | `'MobileNetV1'` |
 | `anchors (list[list[int]])` | 预定义锚框的大小                       | `[[10, 13], [16, 30], [33, 23], [30, 61], [62, 45], [59, 119], [116, 90], [156, 198], [373, 326]]` |

docs/intro/model_cons_params_en.md

@@ -443,6 +443,7 @@ The YOLOv3 implementation based on PaddlePaddle.
 
 | Parameter Name | Description                                                                                                                 | Default Value |
 | --- |-----------------------------------------------------------------------------------------------------------------------------| --- |
+| `rotate (bool)` | If True, the model performs rotated object detection | `False` |
 | `num_classes (int)` | Number of target classes                                                                                                    | `80` |
 | `backbone (str)` | Backbone network to use                                                                                      | `'MobileNetV1'` |
 | `anchors (list[list[int]])` | Sizes of predefined anchor boxes                                                                                                   | `[[10, 13], [16, 30], [33, 23], [30, 61], [62, 45 ], [59, 119], [116, 90], [156, 198], [373, 326]]` |

docs/intro/model_zoo_cn.md

@@ -33,6 +33,7 @@ PaddleRS目前已支持的全部模型如下（标注\*的为遥感专用模型
 | 图像复原 | NAFNet | 是 |
 | 图像复原 | SwinIR | 是 |
 | 目标检测 | Faster R-CNN | 否 |
+| 目标检测 | FCOSR | 否 |
 | 目标检测 | PP-YOLO | 否 |
 | 目标检测 | PP-YOLO Tiny | 否 |
 | 目标检测 | PP-YOLOv2 | 否 |

docs/intro/model_zoo_en.md

@@ -33,6 +33,7 @@ All models currently supported by PaddleRS are listed below (those marked \* are
 | Image Restoration | SwinIR | Yes |
 | Image Restoration | NAFNet | Yes |
 | Object Detection | Faster R-CNN | No |
+| Object Detection | FCOSR | No |
 | Object Detection | PP-YOLO | No |
 | Object Detection | PP-YOLO Tiny | No |
 | Object Detection | PP-YOLOv2 | No |

paddlers/datasets/base.py

@@ -18,15 +18,22 @@
 from paddle.fluid.dataloader.collate import default_collate_fn
 
 from paddlers.utils import get_num_workers
-from paddlers.transforms import construct_sample_from_dict, Compose
+import paddlers.utils.logging as logging
+from paddlers.transforms import construct_sample_from_dict, Compose, BatchCompose
 
 
 class BaseDataset(Dataset):
     _KEYS_TO_KEEP = None
     _KEYS_TO_DISCARD = None
     _collate_trans_info = False
 
-    def __init__(self, data_dir, label_list, transforms, num_workers, shuffle):
+    def __init__(self,
+                 data_dir,
+                 label_list,
+                 transforms,
+                 num_workers,
+                 shuffle,
+                 batch_transforms=None):
         super(BaseDataset, self).__init__()
 
         self.data_dir = data_dir
@@ -37,6 +44,8 @@ def __init__(self, data_dir, label_list, transforms, num_workers, shuffle):
 
         self.num_workers = get_num_workers(num_workers)
         self.shuffle = shuffle
+        self.batch_transforms = None
+        self.build_collate_fn(batch_transforms)
 
     def __getitem__(self, idx):
         sample = construct_sample_from_dict(self.file_list[idx])
@@ -59,8 +68,25 @@ def collate_fn(self, batch):
             for key in self._KEYS_TO_DISCARD:
                 for s, _ in batch:
                     s.pop(key, None)
+
+        samples = [s[0] for s in batch]
+
+        if self.batch_transforms:
+            samples = self.batch_transforms(samples)
+
         if self._collate_trans_info:
-            return default_collate_fn(
-                [s[0] for s in batch]), [s[1] for s in batch]
+            return default_collate_fn(samples), [s[1] for s in batch]
         else:
-            return default_collate_fn([s[0] for s in batch])
+            return default_collate_fn(samples)
+
+    def build_collate_fn(self, batch_transforms, collate_fn_constructor=None):
+        if self.batch_transforms is not None and batch_transforms:
+            logging.warning(
+                "The initial `batch_transforms` will be overwritten.")
+        if batch_transforms is not None:
+            batch_transforms = copy.deepcopy(batch_transforms)
+            if isinstance(batch_transforms, list):
+                batch_transforms = BatchCompose(batch_transforms)
+            self.batch_transforms = batch_transforms
+        if collate_fn_constructor:
+            self.collate_fn = collate_fn_constructor(self)

paddlers/datasets/cd_dataset.py

@@ -55,9 +55,10 @@ def __init__(self,
                  num_workers='auto',
                  shuffle=False,
                  with_seg_labels=False,
-                 binarize_labels=False):
+                 binarize_labels=False,
+                 batch_transforms=None):
         super(CDDataset, self).__init__(data_dir, label_list, transforms,
-                                        num_workers, shuffle)
+                                        num_workers, shuffle, batch_transforms)
 
         DELIMETER = ' '
 

paddlers/datasets/clas_dataset.py

@@ -42,9 +42,11 @@ def __init__(self,
                  transforms,
                  label_list=None,
                  num_workers='auto',
-                 shuffle=False):
+                 shuffle=False,
+                 batch_transforms=None):
         super(ClasDataset, self).__init__(data_dir, label_list, transforms,
-                                          num_workers, shuffle)
+                                          num_workers, shuffle,
+                                          batch_transforms)
         self.file_list = list()
         self.labels = list()