Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Feature] Support more MMOCR models: DBNet++, TextSnake, ABINet, MRCNN #1534

Merged
merged 24 commits into from
Jan 16, 2023
Merged

[Feature] Support more MMOCR models: DBNet++, TextSnake, ABINet, MRCNN #1534

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
24 commits
Select commit Hold shift + click to select a range
fac3e3b
WIP ocr
AllentDan Nov 14, 2022
d2ba8e0
add mrcnn rewrite
AllentDan Nov 17, 2022
fc8c915
add any rewrite for abinet
AllentDan Nov 17, 2022
f2f6d87
export abinet to onnx
AllentDan Nov 18, 2022
25f87e1
fix abinet onnx export
AllentDan Nov 21, 2022
2737500
support abinet to tensorrt static and modify mmocr.yml
AllentDan Dec 14, 2022
771d67c
Merge branch 'dev-1.x' into more-ocr-models
AllentDan Dec 14, 2022
eadb52c
add textsnake and dbnetpp
AllentDan Dec 14, 2022
fd9c0cd
support mrcnn in ORT and TRT
AllentDan Dec 16, 2022
24515ab
add a condition before update data_preprocessor scope
AllentDan Dec 16, 2022
c62bf53
update doc and mmocr.yml
AllentDan Dec 20, 2022
ddcc8d1
add ut
AllentDan Dec 20, 2022
104187c
markdown and simple config
AllentDan Dec 26, 2022
7c3cb53
write build_pytorch_model in child class
AllentDan Dec 26, 2022
d8337ca
update any_default
AllentDan Dec 26, 2022
d55e0a2
Merge branch 'dev-1.x' into more-ocr-models
AllentDan Dec 26, 2022
890b6e1
remove where in abi_language_decoder___get_length__default
AllentDan Dec 26, 2022
b48412d
keep where
AllentDan Dec 27, 2022
6b22ae7
fix UT
AllentDan Dec 27, 2022
802d308
fix UT
AllentDan Dec 27, 2022
43e57bb
fix UT
AllentDan Dec 27, 2022
844ef66
update mmocr.yml and config description
AllentDan Jan 12, 2023
19fceaf
tensorrt-fp32 -> tensorrt
AllentDan Jan 12, 2023
d962059
update doc
AllentDan Jan 13, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 33 additions & 0 deletions configs/mmocr/text-detection/text-detection_mrcnn_onnxruntime_dynamic.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
_base_ = ['./text-detection_static.py', '../../_base_/backends/onnxruntime.py']
onnx_config = dict(
output_names=['dets', 'labels', 'masks'],
dynamic_axes=dict(
input=dict({
0: 'batch',
2: 'height',
3: 'width'
}),
dets=dict({
0: 'batch',
1: 'num_dets'
}),
labels=dict({
0: 'batch',
1: 'num_dets'
}),
masks=dict({
0: 'batch',
1: 'num_dets',
2: 'height',
3: 'width'
})))
codebase_config = dict(
post_processing=dict(
score_threshold=0.05,
confidence_threshold=0.005,
iou_threshold=0.5,
max_output_boxes_per_class=200,
pre_top_k=5000,
keep_top_k=100,
background_label_id=-1,
export_postprocess_mask=False))
2 changes: 2 additions & 0 deletions configs/mmocr/text-detection/text-detection_mrcnn_tensorrt-fp16_dynamic-320x320-2240x2240.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
_base_ = ['./text-detection_mrcnn_tensorrt_dynamic-320x320-2240x2240.py']
backend_config = dict(common_config=dict(fp16_mode=True))
5 changes: 5 additions & 0 deletions configs/mmocr/text-detection/text-detection_mrcnn_tensorrt-int8_dynamic-320x320-2240x2240.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
_base_ = ['./text-detection_mrcnn_tensorrt_dynamic-320x320-2240x2240.py']

backend_config = dict(common_config=dict(fp16_mode=True, int8_mode=True))

calib_config = dict(create_calib=True, calib_file='calib_data.h5')
45 changes: 45 additions & 0 deletions configs/mmocr/text-detection/text-detection_mrcnn_tensorrt_dynamic-320x320-2240x2240.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
_base_ = ['./text-detection_static.py', '../../_base_/backends/tensorrt.py']
onnx_config = dict(
output_names=['dets', 'labels', 'masks'],
dynamic_axes=dict(
input=dict({
0: 'batch',
2: 'height',
3: 'width'
}),
dets=dict({
0: 'batch',
1: 'num_dets'
}),
labels=dict({
0: 'batch',
1: 'num_dets'
}),
masks=dict({
0: 'batch',
1: 'num_dets',
2: 'height',
3: 'width'
})))

backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 320, 320],
opt_shape=[1, 3, 600, 800],
max_shape=[1, 3, 2240, 2240])))
])

codebase_config = dict(
post_processing=dict(
score_threshold=0.05,
confidence_threshold=0.005,
iou_threshold=0.5,
max_output_boxes_per_class=200,
pre_top_k=5000,
keep_top_k=100,
background_label_id=-1,
export_postprocess_mask=False))
18 changes: 18 additions & 0 deletions configs/mmocr/text-detection/text-detection_mrcnn_torchscript.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
_base_ = [
'../../_base_/torchscript_config.py',
'../../_base_/backends/torchscript.py'
]

ir_config = dict(input_shape=None, output_names=['dets', 'labels', 'masks'])
codebase_config = dict(
type='mmocr',
task='TextDetection',
post_processing=dict(
score_threshold=0.05,
confidence_threshold=0.005,
iou_threshold=0.5,
max_output_boxes_per_class=200,
pre_top_k=5000,
keep_top_k=100,
background_label_id=-1,
export_postprocess_mask=False))
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-fp16_dynamic-1x32x32-1x32x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 1 channel input for CRNN models
_base_ = [
'./text-recognition_dynamic.py', '../../_base_/backends/tensorrt-fp16.py'
]
Expand Down
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-fp16_dynamic-32x32-32x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 3 channel and 32 height input for SATRN models
_base_ = [
'./text-recognition_dynamic.py', '../../_base_/backends/tensorrt-fp16.py'
]
Expand Down
14 changes: 14 additions & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-fp16_dynamic-48x64-48x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
# 3 channel and 48 height for SAR models
_base_ = [
'./text-recognition_dynamic.py', '../../_base_/backends/tensorrt-fp16.py'
]
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 48, 64],
opt_shape=[1, 3, 48, 64],
max_shape=[1, 3, 48, 640])))
])
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-fp16_static-1x32x32.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 1 channel input for CRNN models
_base_ = [
'./text-recognition_static.py', '../../_base_/backends/tensorrt-fp16.py'
]
Expand Down
16 changes: 16 additions & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-fp16_static-32x128.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
# ABINet models use static input 32x128
_base_ = [
'./text-recognition_static.py', '../../_base_/backends/tensorrt-fp16.py'
]

onnx_config = dict(input_shape=[128, 32])
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 32, 128],
opt_shape=[1, 3, 32, 128],
max_shape=[1, 3, 32, 128])))
])
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-int8_dynamic-1x32x32-1x32x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 1 channel input for CRNN models
_base_ = [
'./text-recognition_dynamic.py', '../../_base_/backends/tensorrt-int8.py'
]
Expand Down
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-int8_dynamic-32x32-32x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 3 channel and 32 height input for SATRN models
_base_ = [
'./text-recognition_dynamic.py', '../../_base_/backends/tensorrt-int8.py'
]
Expand Down
14 changes: 14 additions & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-int8_dynamic-48x64-48x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
# 3 channel and 48 height for SAR models
_base_ = [
'./text-recognition_dynamic.py', '../../_base_/backends/tensorrt-int8.py'
]
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 48, 64],
opt_shape=[1, 3, 48, 64],
max_shape=[1, 3, 48, 640])))
])
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-int8_static-1x32x32.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 1 channel input for CRNN models
_base_ = [
'./text-recognition_static.py', '../../_base_/backends/tensorrt-int8.py'
]
Expand Down
16 changes: 16 additions & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt-int8_static-32x128.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
# ABINet models use static input 32x128
_base_ = [
'./text-recognition_static.py', '../../_base_/backends/tensorrt-int8.py'
]

onnx_config = dict(input_shape=[128, 32])
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 32, 128],
opt_shape=[1, 3, 32, 128],
max_shape=[1, 3, 32, 128])))
])
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt_dynamic-1x32x32-1x32x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 1 channel input for CRNN models
_base_ = ['./text-recognition_dynamic.py', '../../_base_/backends/tensorrt.py']
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
Expand Down
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt_dynamic-32x32-32x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 3 channel and 32 height input for SATRN models
_base_ = ['./text-recognition_dynamic.py', '../../_base_/backends/tensorrt.py']
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
Expand Down
12 changes: 12 additions & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt_dynamic-48x64-48x640.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
# 3 channel and 48 height for SAR models
_base_ = ['./text-recognition_dynamic.py', '../../_base_/backends/tensorrt.py']
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 48, 64],
opt_shape=[1, 3, 48, 64],
max_shape=[1, 3, 48, 640])))
])
1 change: 1 addition & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt_static-1x32x32.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
# 1 channel input for CRNN models
_base_ = ['./text-recognition_static.py', '../../_base_/backends/tensorrt.py']

onnx_config = dict(input_shape=[32, 32])
Expand Down
14 changes: 14 additions & 0 deletions configs/mmocr/text-recognition/text-recognition_tensorrt_static-32x128.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
# ABINet models use static input 32x128
_base_ = ['./text-recognition_static.py', '../../_base_/backends/tensorrt.py']

onnx_config = dict(input_shape=[128, 32])
backend_config = dict(
common_config=dict(max_workspace_size=1 << 30),
model_inputs=[
dict(
input_shapes=dict(
input=dict(
min_shape=[1, 3, 32, 128],
opt_shape=[1, 3, 32, 128],
max_shape=[1, 3, 32, 128])))
])
122 changes: 122 additions & 0 deletions docs/en/03-benchmark/benchmark.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1178,6 +1178,42 @@ Users can directly test the performance through [how_to_evaluate_a_model.md](../
<td align="center">0.7949</td>
<td align="center">0.7950</td>
</tr>
<tr>
<td align="center" rowspan="3"><a href="https://github.com/open-mmlab/mmocr/blob/1.x/configs/textdet/dbnetpp/dbnetpp_resnet50_fpnc_1200e_icdar2015.py">DBNetpp</a></td>
<td align="center" rowspan="3">TextDetection</td>
<td align="center" rowspan="3">ICDAR2015</td>
<td align="center">recall</td>
<td align="center">0.8209</td>
<td align="center">0.8209</td>
<td align="center">0.8209</td>
<td align="center">0.8199</td>
<td align="center">0.8204</td>
<td align="center">0.8204</td>
<td align="center">-</td>
<td align="center">0.8209</td>
</tr>
<tr>
<td align="center">precision</td>
<td align="center">0.9079</td>
<td align="center">0.9079</td>
<td align="center">0.9079</td>
<td align="center">0.9117</td>
<td align="center">0.9117</td>
<td align="center">0.9142</td>
<td align="center">-</td>
<td align="center">0.9079</td>
</tr>
<tr>
<td align="center">hmean</td>
<td align="center">0.8622</td>
<td align="center">0.8622</td>
<td align="center">0.8622</td>
<td align="center">0.8634</td>
<td align="center">0.8637</td>
<td align="center">0.8648</td>
<td align="center">-</td>
<td align="center">0.8622</td>
</tr>
<tr>
<td align="center" rowspan="3"><a href="https://github.com/open-mmlab/mmocr/blob/1.x/configs/textdet/psenet/psenet_resnet50_fpnf_600e_icdar2015.py">PSENet</a></td>
<td align="center" rowspan="3">TextDetection</td>
Expand Down Expand Up @@ -1250,6 +1286,78 @@ Users can directly test the performance through [how_to_evaluate_a_model.md](../
<td align="center">-</td>
<td align="center">0.7955</td>
</tr>
<tr>
<td align="center" rowspan="3"><a href="https://github.com/open-mmlab/mmocr/blob/1.x/configs/textdet/textsnake/textsnake_resnet50_fpn-unet_1200e_ctw1500.py">TextSnake</a></td>
<td align="center" rowspan="3">TextDetection</td>
<td align="center" rowspan="3">CTW1500</td>
<td align="center">recall</td>
<td align="center">0.8052</td>
<td align="center">0.8052</td>
<td align="center">0.8052</td>
<td align="center">0.8055</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center">precision</td>
<td align="center">0.8535</td>
<td align="center">0.8535</td>
<td align="center">0.8535</td>
<td align="center">0.8538</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center">hmean</td>
<td align="center">0.8286</td>
<td align="center">0.8286</td>
<td align="center">0.8286</td>
<td align="center">0.8290</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center" rowspan="3"><a href="https://github.com/open-mmlab/mmocr/blob/1.x/configs/textdet/maskrcnn/mask-rcnn_resnet50_fpn_160e_icdar2015.py">MaskRCNN</a></td>
<td align="center" rowspan="3">TextDetection</td>
<td align="center" rowspan="3">ICDAR2015</td>
<td align="center">recall</td>
<td align="center">0.7766</td>
<td align="center">0.7766</td>
<td align="center">0.7766</td>
<td align="center">0.7766</td>
<td align="center">0.7761</td>
<td align="center">0.7670</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center">precision</td>
<td align="center">0.8644</td>
<td align="center">0.8644</td>
<td align="center">0.8644</td>
<td align="center">0.8644</td>
<td align="center">0.8630</td>
<td align="center">0.8705</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center">hmean</td>
<td align="center">0.8182</td>
<td align="center">0.8182</td>
<td align="center">0.8182</td>
<td align="center">0.8182</td>
<td align="center">0.8172</td>
<td align="center">0.8155</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center"><a href="https://github.com/open-mmlab/mmocr/blob/1.x/configs/textrecog/crnn/crnn_mini-vgg_5e_mj.py">CRNN</a></td>
<td align="center">TextRecognition</td>
Expand Down Expand Up @@ -1292,6 +1400,20 @@ Users can directly test the performance through [how_to_evaluate_a_model.md](../
<td align="center">-</td>
<td align="center">-</td>
</tr>
<tr>
<td align="center"><a href="https://github.com/open-mmlab/mmocr/blob/1.x/configs/textrecog/abinet/abinet_20e_st-an_mj.py">ABINet</a></td>
<td align="center">TextRecognition</td>
<td align="center">IIIT5K</td>
<td align="center">acc</td>
<td align="center">0.9603</td>
<td align="center">0.9563</td>
<td align="center">0.9563</td>
<td align="center">0.9573</td>
<td align="center">0.9507</td>
<td align="center">0.9510</td>
<td align="center">-</td>
<td align="center">-</td>
</tr>
</tbody>
</table>
</div>
Expand Down
Loading