[WIP] Support ICDAR dataset.

configs/_base_/datasets/icdar.py

@@ -0,0 +1,47 @@
+# dataset settings
+dataset_type = 'IcdarDataset'
+data_root = 'work_dirs/icdar2015'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(type='RResize', img_scale=(1000, 800)),
+    dict(type='RRandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(1000, 800),
+        flip=False,
+        transforms=[
+            dict(type='RResize'),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='Pad', size_divisor=32),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=f'{data_root}/instances_training.json',
+        img_prefix=f'{data_root}/imgs',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        ann_file=f'{data_root}/instances_test.json',
+        img_prefix=f'{data_root}/imgs',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=f'{data_root}/instances_test.json',
+        img_prefix=f'{data_root}/imgs',
+        pipeline=test_pipeline))

configs/s2anet/s2anet_r50_fpn_1x_icdar2015_le135.py

@@ -0,0 +1,123 @@
+_base_ = [
+    '../_base_/datasets/icdar.py', '../_base_/schedules/schedule_1x.py',
+    '../_base_/default_runtime.py'
+]
+
+angle_version = 'le135'
+model = dict(
+    type='S2ANet',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        zero_init_residual=False,
+        norm_cfg=dict(type='BN', requires_grad=True),
+        norm_eval=True,
+        style='pytorch',
+        init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        start_level=1,
+        add_extra_convs='on_input',
+        num_outs=5),
+    fam_head=dict(
+        type='RotatedRetinaHead',
+        num_classes=1,
+        in_channels=256,
+        stacked_convs=2,
+        feat_channels=256,
+        assign_by_circumhbbox=None,
+        anchor_generator=dict(
+            type='RotatedAnchorGenerator',
+            scales=[4],
+            ratios=[1.0],
+            strides=[8, 16, 32, 64, 128]),
+        bbox_coder=dict(
+            type='DeltaXYWHAOBBoxCoder',
+            angle_range=angle_version,
+            norm_factor=1,
+            edge_swap=False,
+            proj_xy=True,
+            target_means=(.0, .0, .0, .0, .0),
+            target_stds=(1.0, 1.0, 1.0, 1.0, 1.0)),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=1.0),
+        loss_bbox=dict(type='SmoothL1Loss', beta=0.11, loss_weight=1.0)),
+    align_cfgs=dict(
+        type='AlignConv',
+        kernel_size=3,
+        channels=256,
+        featmap_strides=[8, 16, 32, 64, 128]),
+    odm_head=dict(
+        type='ODMRefineHead',
+        num_classes=1,
+        in_channels=256,
+        stacked_convs=2,
+        feat_channels=256,
+        assign_by_circumhbbox=None,
+        anchor_generator=dict(
+            type='PseudoAnchorGenerator', strides=[8, 16, 32, 64, 128]),
+        bbox_coder=dict(
+            type='DeltaXYWHAOBBoxCoder',
+            angle_range=angle_version,
+            norm_factor=1,
+            edge_swap=False,
+            proj_xy=True,
+            target_means=(0.0, 0.0, 0.0, 0.0, 0.0),
+            target_stds=(1.0, 1.0, 1.0, 1.0, 1.0)),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=1.0),
+        loss_bbox=dict(type='SmoothL1Loss', beta=0.11, loss_weight=1.0)),
+    train_cfg=dict(
+        fam_cfg=dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.5,
+                neg_iou_thr=0.4,
+                min_pos_iou=0,
+                ignore_iof_thr=-1,
+                iou_calculator=dict(type='RBboxOverlaps2D')),
+            allowed_border=-1,
+            pos_weight=-1,
+            debug=False),
+        odm_cfg=dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.5,
+                neg_iou_thr=0.4,
+                min_pos_iou=0,
+                ignore_iof_thr=-1,
+                iou_calculator=dict(type='RBboxOverlaps2D')),
+            allowed_border=-1,
+            pos_weight=-1,
+            debug=False)),
+    test_cfg=dict(
+        nms_pre=2000,
+        min_bbox_size=0,
+        score_thr=0.05,
+        nms=dict(iou_thr=0.7),
+        max_per_img=1000))
+
+# evaluation
+evaluation = dict(interval=1, metric='mAP')
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[24, 32, 38])
+runner = dict(type='EpochBasedRunner', max_epochs=40)
+checkpoint_config = dict(interval=10)

mmrotate/datasets/__init__.py

@@ -1,7 +1,8 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from .builder import build_dataset  # noqa: F401, F403
 from .dota import DOTADataset  # noqa: F401, F403
+from .icdar import IcdarDataset  # noqa: F401, F403
 from .pipelines import *  # noqa: F401, F403
 from .sar import SARDataset  # noqa: F401, F403
 
-__all__ = ['SARDataset', 'DOTADataset', 'build_dataset']
+__all__ = ['SARDataset', 'DOTADataset', 'IcdarDataset', 'build_dataset']

mmrotate/datasets/icdar.py

@@ -0,0 +1,94 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import numpy as np
+from mmdet.datasets.api_wrappers import COCO
+
+from mmrotate.core import poly2obb_np
+from .builder import ROTATED_DATASETS
+from .dota import DOTADataset
+
+
+@ROTATED_DATASETS.register_module()
+class IcdarDataset(DOTADataset):
+    """ICDAR text dataset for rotated object detection (Support ICDAR2015 and
+    ICDAR2017)."""
+    CLASSES = ('text', )
+
+    def __init__(self,
+                 ann_file,
+                 pipeline,
+                 version='oc',
+                 select_first_k=-1,
+                 **kwargs):
+        self.version = version
+        self.select_first_k = select_first_k
+
+        super().__init__(ann_file, pipeline, **kwargs)
+
+    def load_annotations(self, ann_file):
+        """Load annotation from COCO style annotation file.
+
+        Args:
+            ann_file (str): Path of annotation file.
+        Returns:
+            list[dict]: Annotation info.
+        """
+        self.coco = COCO(ann_file)
+        self.cat_ids = self.coco.get_cat_ids(cat_names=self.CLASSES)
+        self.cat2label = {cat_id: i for i, cat_id in enumerate(self.cat_ids)}
+        self.img_ids = self.coco.get_img_ids()
+        data_infos = []
+        count = 0
+        for i in self.img_ids:
+            data_info = {}
+            info = self.coco.load_imgs([i])[0]
+            data_info['filename'] = info['file_name']
+            data_info['ann'] = {}
+            img_id = info['id']
+            ann_ids = self.coco.get_ann_ids(img_ids=[img_id])
+            ann_info = self.coco.load_anns(ann_ids)
+            gt_bboxes = []
+            gt_labels = []
+            gt_polygons = []
+            for ann in ann_info:
+                if ann.get('ignore', False):
+                    continue
+                x1, y1, w, h = ann['bbox']
+                if ann['area'] <= 0 or w < 1 or h < 1:
+                    continue
+                if ann['category_id'] not in self.cat_ids:
+                    continue
+                try:
+                    x, y, w, h, a = poly2obb_np(ann['segmentation'][0],
+                                                self.version)
+                except:  # noqa: E722
+                    continue
+                gt_bboxes.append([x, y, w, h, a])
+                gt_labels.append(ann['category_id'])
+                gt_polygons.append(ann['segmentation'][0])
+
+                if gt_bboxes:
+                    data_info['ann']['bboxes'] = np.array(
+                        gt_bboxes, dtype=np.float32)
+                    data_info['ann']['labels'] = np.array(
+                        gt_labels, dtype=np.int64)
+                    data_info['ann']['polygons'] = np.array(
+                        gt_polygons, dtype=np.float32)
+                else:
+                    data_info['ann']['bboxes'] = np.zeros((0, 5),
+                                                          dtype=np.float32)
+                    data_info['ann']['labels'] = np.array([], dtype=np.int64)
+                    data_info['ann']['polygons'] = np.zeros((0, 8),
+                                                            dtype=np.float32)
+                data_info['ann']['bboxes_ignore'] = np.zeros(
+                    (0, 5), dtype=np.float32)
+                data_info['ann']['labels_ignore'] = np.array(
+                    [], dtype=np.int64)
+                data_info['ann']['polygons_ignore'] = np.zeros(
+                    (0, 8), dtype=np.float32)
+
+                data_infos.append(data_info)
+            count = count + 1
+            if count > self.select_first_k and self.select_first_k > 0:
+                break
+
+        return data_infos

requirements/runtime.txt

@@ -4,6 +4,7 @@ mmcv-full
 mmdet
 numpy
 pycocotools
+shapely
 six
 terminaltables
 torch

setup.cfg

@@ -3,7 +3,7 @@ line_length = 79
 multi_line_output = 0
 known_standard_library = setuptools
 known_first_party = mmrotate
-known_third_party = PIL,cv2,e2cnn,matplotlib,mmcv,mmdet,numpy,pytest,pytorch_sphinx_theme,terminaltables,torch
+known_third_party = PIL,cv2,e2cnn,matplotlib,mmcv,mmdet,numpy,pytest,pytorch_sphinx_theme,shapely,terminaltables,torch
 no_lines_before = STDLIB,LOCALFOLDER
 default_section = THIRDPARTY
 

tools/data/dota/split/img_split.py

@@ -235,9 +235,6 @@ def bbox_overlaps_iof(bboxes1, bboxes2, eps=1e-6):
 
     l, t, r, b = [bboxes2[..., i] for i in range(4)]
     polys2 = np.stack([l, t, r, t, r, b, l, b], axis=-1)
-    if shgeo is None:
-        raise ImportError('Please run "pip install shapely" '
-                          'to install shapely first.')
     sg_polys1 = [shgeo.Polygon(p) for p in bboxes1.reshape(rows, -1, 2)]
     sg_polys2 = [shgeo.Polygon(p) for p in polys2.reshape(cols, -1, 2)]
     overlaps = np.zeros(h_overlaps.shape)

tools/data/icdar/README.md

@@ -0,0 +1,47 @@
+# ICDAR Dataset
+
+<!-- [DATASET] -->
+
+## Overview
+
+The data structure is as follows:
+```text
+├── icdar2015
+│   ├── imgs
+│   ├── instances_test.json
+│   └── instances_training.json
+├── icdar2017
+│   ├── imgs
+│   ├── instances_training.json
+│   └── instances_val.json
+```
+
+|Dataset|Images|                                                                                      |  Annotation Files                                                                                                      |                         |                                                                                                |
+| :-------: | :------------------------------------------------------------: | :----------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------: | :-------------------------------------: | :--------------------------------------------------------------------------------------------: |
+|      |                                                                                      |                                                training                                                |               validation                |                                            testing                                             |       |
+| ICDAR2015 | [homepage](https://rrc.cvc.uab.es/?ch=4&com=downloads)     | [instances_training.json](https://download.openmmlab.com/mmocr/data/icdar2015/instances_training.json) |                    -                    | [instances_test.json](https://download.openmmlab.com/mmocr/data/icdar2015/instances_test.json) |
+| ICDAR2017 | [homepage](https://rrc.cvc.uab.es/?ch=8&com=downloads)     | [instances_training.json](https://download.openmmlab.com/mmocr/data/icdar2017/instances_training.json) | [instances_val.json](https://download.openmmlab.com/mmocr/data/icdar2017/instances_val.json) | - |       |       |
+
+
+## Preparation Steps
+### ICDAR 2015
+- Step1: Download `ch4_training_images.zip`, `ch4_test_images.zip`, `ch4_training_localization_transcription_gt.zip`, `Challenge4_Test_Task1_GT.zip` from [homepage](https://rrc.cvc.uab.es/?ch=4&com=downloads)
+- Step2:
+```bash
+mkdir icdar2015 && cd icdar2015
+mkdir imgs && mkdir annotations
+# For images,
+mv ch4_training_images imgs/training
+mv ch4_test_images imgs/test
+# For annotations,
+mv ch4_training_localization_transcription_gt annotations/training
+mv Challenge4_Test_Task1_GT annotations/test
+```
+- Step3: Download [instances_training.json](https://download.openmmlab.com/mmocr/data/icdar2015/instances_training.json) and [instances_test.json](https://download.openmmlab.com/mmocr/data/icdar2015/instances_test.json) and move them to `icdar2015`
+- Or, generate `instances_training.json` and `instances_test.json` with following command:
+```bash
+python tools/data/icdar/icdar_converter.py /path/to/icdar2015 -o /path/to/icdar2015 -d icdar2015 --split-list training test
+```
+
+### ICDAR 2017
+- Follow similar steps as [ICDAR 2015](#icdar-2015).