support batch inference during testing (#310)

* support batch inference during testing * fix unittest * update docs using url * set cfg for train, val and test * update docs * update docs and test.py * samples_per_gpu as global setting * changes revert
open-mmlab · Jun 23, 2021 · 82f64a5 · 82f64a5
1 parent e6cb750
commit 82f64a5
Show file tree

Hide file tree

Showing 30 changed files with 263 additions and 95 deletions.
diff --git a/configs/textdet/dbnet/dbnet_r18_fpnc_1200e_icdar2015.py b/configs/textdet/dbnet/dbnet_r18_fpnc_1200e_icdar2015.py
@@ -74,6 +74,8 @@
 data = dict(
     samples_per_gpu=16,
     workers_per_gpu=8,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/dbnet/dbnet_r50dcnv2_fpnc_1200e_icdar2015.py b/configs/textdet/dbnet/dbnet_r50dcnv2_fpnc_1200e_icdar2015.py
@@ -83,6 +83,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/drrg/drrg_r50_fpn_unet_1200e_ctw1500.py b/configs/textdet/drrg/drrg_r50_fpn_unet_1200e_ctw1500.py
@@ -91,6 +91,8 @@
 data = dict(
     samples_per_gpu=4,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=f'{data_root}/instances_training.json',

diff --git a/configs/textdet/fcenet/fcenet_r50_fpn_1500e_icdar2015.py b/configs/textdet/fcenet/fcenet_r50_fpn_1500e_icdar2015.py
@@ -96,6 +96,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/fcenet/fcenet_r50dcnv2_fpn_1500e_ctw1500.py b/configs/textdet/fcenet/fcenet_r50dcnv2_fpn_1500e_ctw1500.py
@@ -95,6 +95,8 @@
 data = dict(
     samples_per_gpu=6,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/maskrcnn/mask_rcnn_r50_fpn_160e_ctw1500.py b/configs/textdet/maskrcnn/mask_rcnn_r50_fpn_160e_ctw1500.py
@@ -47,6 +47,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/maskrcnn/mask_rcnn_r50_fpn_160e_icdar2015.py b/configs/textdet/maskrcnn/mask_rcnn_r50_fpn_160e_icdar2015.py
@@ -46,6 +46,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/maskrcnn/mask_rcnn_r50_fpn_160e_icdar2017.py b/configs/textdet/maskrcnn/mask_rcnn_r50_fpn_160e_icdar2017.py
@@ -47,6 +47,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/panet/panet_r18_fpem_ffm_600e_ctw1500.py b/configs/textdet/panet/panet_r18_fpem_ffm_600e_ctw1500.py
@@ -82,6 +82,8 @@
 data = dict(
     samples_per_gpu=2,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/panet/panet_r18_fpem_ffm_600e_icdar2015.py b/configs/textdet/panet/panet_r18_fpem_ffm_600e_icdar2015.py
@@ -80,6 +80,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/panet/panet_r50_fpem_ffm_600e_icdar2017.py b/configs/textdet/panet/panet_r50_fpem_ffm_600e_icdar2017.py
@@ -75,6 +75,8 @@
 data = dict(
     samples_per_gpu=4,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/psenet/psenet_r50_fpnf_600e_ctw1500.py b/configs/textdet/psenet/psenet_r50_fpnf_600e_ctw1500.py
@@ -89,6 +89,8 @@
 data = dict(
     samples_per_gpu=2,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/psenet/psenet_r50_fpnf_600e_icdar2015.py b/configs/textdet/psenet/psenet_r50_fpnf_600e_icdar2015.py
@@ -89,6 +89,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/psenet/psenet_r50_fpnf_600e_icdar2017.py b/configs/textdet/psenet/psenet_r50_fpnf_600e_icdar2017.py
@@ -84,6 +84,8 @@
 data = dict(
     samples_per_gpu=8,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=data_root + '/instances_training.json',

diff --git a/configs/textdet/textsnake/textsnake_r50_fpn_unet_1200e_ctw1500.py b/configs/textdet/textsnake/textsnake_r50_fpn_unet_1200e_ctw1500.py
@@ -94,6 +94,8 @@
 data = dict(
     samples_per_gpu=4,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type=dataset_type,
         ann_file=f'{data_root}/instances_training.json',

diff --git a/configs/textrecog/crnn/crnn_academic_dataset.py b/configs/textrecog/crnn/crnn_academic_dataset.py
@@ -148,6 +148,8 @@
 data = dict(
     samples_per_gpu=64,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(type='ConcatDataset', datasets=[train1]),
     val=dict(
         type='ConcatDataset',

diff --git a/configs/textrecog/nrtr/nrtr_r31_1by16_1by8_academic.py b/configs/textrecog/nrtr/nrtr_r31_1by16_1by8_academic.py
@@ -152,6 +152,8 @@
 data = dict(
     samples_per_gpu=128,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(type='ConcatDataset', datasets=[train1, train2]),
     val=dict(
         type='ConcatDataset',

diff --git a/configs/textrecog/nrtr/nrtr_r31_1by8_1by4_academic.py b/configs/textrecog/nrtr/nrtr_r31_1by8_1by4_academic.py
@@ -152,6 +152,8 @@
 data = dict(
     samples_per_gpu=128,
     workers_per_gpu=4,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(type='ConcatDataset', datasets=[train1, train2]),
     val=dict(
         type='ConcatDataset',

diff --git a/configs/textrecog/robust_scanner/robustscanner_r31_academic.py b/configs/textrecog/robust_scanner/robustscanner_r31_academic.py
@@ -182,6 +182,8 @@
 data = dict(
     samples_per_gpu=64,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type='ConcatDataset',
         datasets=[

diff --git a/configs/textrecog/sar/sar_r31_parallel_decoder_academic.py b/configs/textrecog/sar/sar_r31_parallel_decoder_academic.py
@@ -204,6 +204,8 @@
 data = dict(
     samples_per_gpu=64,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type='ConcatDataset',
         datasets=[

diff --git a/configs/textrecog/sar/sar_r31_parallel_decoder_chinese.py b/configs/textrecog/sar/sar_r31_parallel_decoder_chinese.py
@@ -119,6 +119,8 @@
 data = dict(
     samples_per_gpu=40,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(type='ConcatDataset', datasets=[train]),
     val=dict(type='ConcatDataset', datasets=[test]),
     test=dict(type='ConcatDataset', datasets=[test]))

diff --git a/configs/textrecog/sar/sar_r31_parallel_decoder_toy_dataset.py b/configs/textrecog/sar/sar_r31_parallel_decoder_toy_dataset.py
@@ -30,24 +30,19 @@
 test_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(
-        type='MultiRotateAugOCR',
-        rotate_degrees=[0, 90, 270],
-        transforms=[
-            dict(
-                type='ResizeOCR',
-                height=48,
-                min_width=48,
-                max_width=160,
-                keep_aspect_ratio=True),
-            dict(type='ToTensorOCR'),
-            dict(type='NormalizeOCR', **img_norm_cfg),
-            dict(
-                type='Collect',
-                keys=['img'],
-                meta_keys=[
-                    'filename', 'ori_shape', 'img_shape', 'valid_ratio',
-                    'img_norm_cfg', 'ori_filename'
-                ]),
+        type='ResizeOCR',
+        height=48,
+        min_width=48,
+        max_width=160,
+        keep_aspect_ratio=True),
+    dict(type='ToTensorOCR'),
+    dict(type='NormalizeOCR', **img_norm_cfg),
+    dict(
+        type='Collect',
+        keys=['img'],
+        meta_keys=[
+            'filename', 'ori_shape', 'img_shape', 'valid_ratio',
+            'img_norm_cfg', 'ori_filename'
         ])
 ]
 
@@ -66,7 +61,7 @@
             keys=['filename', 'text'],
             keys_idx=[0, 1],
             separator=' ')),
-    pipeline=train_pipeline,
+    pipeline=None,
     test_mode=False)
 
 train_anno_file2 = 'tests/data/ocr_toy_dataset/label.lmdb'
@@ -82,7 +77,7 @@
             keys=['filename', 'text'],
             keys_idx=[0, 1],
             separator=' ')),
-    pipeline=train_pipeline,
+    pipeline=None,
     test_mode=False)
 
 test_anno_file1 = 'tests/data/ocr_toy_dataset/label.lmdb'
@@ -92,20 +87,25 @@
     ann_file=test_anno_file1,
     loader=dict(
         type='LmdbLoader',
-        repeat=1,
+        repeat=10,
         parser=dict(
             type='LineStrParser',
             keys=['filename', 'text'],
             keys_idx=[0, 1],
             separator=' ')),
-    pipeline=test_pipeline,
+    pipeline=None,
     test_mode=True)
 
 data = dict(
-    samples_per_gpu=16,
     workers_per_gpu=2,
-    train=dict(type='ConcatDataset', datasets=[train1, train2]),
-    val=dict(type='ConcatDataset', datasets=[test]),
-    test=dict(type='ConcatDataset', datasets=[test]))
+    samples_per_gpu=8,
+    train=dict(
+        type='UniformConcatDataset',
+        datasets=[train1, train2],
+        pipeline=train_pipeline),
+    val=dict(
+        type='UniformConcatDataset', datasets=[test], pipeline=test_pipeline),
+    test=dict(
+        type='UniformConcatDataset', datasets=[test], pipeline=test_pipeline))
 
 evaluation = dict(interval=1, metric='acc')
diff --git a/configs/textrecog/sar/sar_r31_sequential_decoder_academic.py b/configs/textrecog/sar/sar_r31_sequential_decoder_academic.py
@@ -204,6 +204,8 @@
 data = dict(
     samples_per_gpu=64,
     workers_per_gpu=2,
+    val_dataloader=dict(samples_per_gpu=1),
+    test_dataloader=dict(samples_per_gpu=1),
     train=dict(
         type='ConcatDataset',
         datasets=[

diff --git a/docs/getting_started.md b/docs/getting_started.md
@@ -202,7 +202,12 @@ To support the tasks of `text detection`, `text recognition` and `key informatio
 
 Here we show some examples of using different combination of `loader` and `parser`.
 
-#### Encoder-Decoder-Based Text Recognition Task
+#### Text Recognition Task
+
+##### OCRDataset
+
+<small>*Dataset for encoder-decoder based recognizer*</small>
+
 ```python
 dataset_type = 'OCRDataset'
 img_prefix = 'tests/data/ocr_toy_dataset/imgs'
@@ -225,7 +230,7 @@ train = dict(
 You can check the content of the annotation file in `tests/data/ocr_toy_dataset/label.txt`.
 The combination of `HardDiskLoader` and `LineStrParser` will return a dict for each file by calling `__getitem__`: `{'filename': '1223731.jpg', 'text': 'GRAND'}`.
 
-##### Optional Arguments:
+**Optional Arguments:**
 
 - `repeat`: The number of repeated lines in the annotation files. For example, if there are `10` lines in the annotation file, setting `repeat=10` will generate a corresponding annotation file with size `100`.
 
@@ -254,7 +259,10 @@ train = dict(
     test_mode=False)
 ```
 
-#### Segmentation-Based Text Recognition Task
+##### OCRSegDataset
+
+<small>*Dataset for segmentation-based recognizer*</small>
+
 ```python
 prefix = 'tests/data/ocr_char_ann_toy_dataset/'
 train = dict(
@@ -277,6 +285,11 @@ The combination of `HardDiskLoader` and `LineJsonParser` will return a dict for
 ```
 
 #### Text Detection Task
+
+##### TextDetDataset
+
+<small>*Dataset with annotation file in line-json txt format*</small>
+
 ```python
 dataset_type = 'TextDetDataset'
 img_prefix = 'tests/data/toy_dataset/imgs'
@@ -302,7 +315,10 @@ The combination of `HardDiskLoader` and `LineJsonParser` will return a dict for
 ```
 
 
-### COCO-like Dataset
+##### IcdarDataset
+
+<small>*Dataset with annotation file in coco-like json format*</small>
+
 For text detection, you can also use an annotation file in a COCO format that is defined in [mmdet](https://github.com/open-mmlab/mmdetection/blob/master/mmdet/datasets/coco.py):
 ```python
 dataset_type = 'IcdarDataset'
@@ -325,4 +341,40 @@ You can check the content of the annotation file in `tests/data/toy_dataset/inst
   ```shell
   python tools/data_converter/ctw1500_converter.py ${src_root_path} -o ${out_path} --split-list training test
   ```
+
+#### UniformConcatDataset
+
+To use the `universal pipeline` for multiple datasets, we design `UniformConcatDataset`.
+For example, apply `train_pipeline` for both `train1` and `train2`,
+
+```python
+data = dict(
+    ...
+    train=dict(
+        type='UniformConcatDataset',
+        datasets=[train1, train2],
+        pipeline=train_pipeline))
+```
+
+Meanwhile, we have
+- train_dataloader
+- val_dataloader
+- test_dataloader
+
+to give specific settings. They will override the general settings in `data` dict.
+For example,
+
+```python
+data = dict(
+    workers_per_gpu=2,                                          # global setting
+    train_dataloader=dict(samples_per_gpu=8, drop_last=True),   # train-specific setting
+    val_dataloader=dict(samples_per_gpu=8, workers_per_gpu=1),  # val-specific setting
+    test_dataloader=dict(samples_per_gpu=8),                    # test-specific setting
+    ...
 ```
+`workers_per_gpu` is global setting and `train_dataloader` and `val_dataloader` will inherit the values.
+`val_dataloader` override the value by `workers_per_gpu=1`.
+
+To activate `batch inference` for `val` and `test`, please set `val_dataloader=dict(samples_per_gpu=8)` and `test_dataloader=dict(samples_per_gpu=8)` as above.
+Or just set `samples_per_gpu=8` as global setting.
+See [config](/configs/textrecog/sar/sar_r31_parallel_decoder_toy_dataset.py) for an example.