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

Jump to bottom

update rknn-toolkit2 version to 1.5 #2278

Merged
merged 5 commits into from
Jul 19, 2023
Merged

update rknn-toolkit2 version to 1.5 #2278

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
8576500
update rknn-toolkit2 version to 1.5
AllentDan Jul 14, 2023
dfab479
fix lint
AllentDan Jul 14, 2023
74305ce
cmake ..
AllentDan Jul 17, 2023
cd11954
fix int type
AllentDan Jul 17, 2023
aa22144
fix
AllentDan Jul 17, 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
2 changes: 1 addition & 1 deletion csrc/mmdeploy/net/rknn/rknn_net.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -170,7 +170,7 @@ Result<void> RKNNNet::Init(const Value& args) {
MMDEPLOY_ERROR("rknn query 'RKNN_QUERY_INPUT_ATTR' fail! ret: {}", ret);
return Status(eFail);
}
if (attr.type != RKNN_TENSOR_UINT8) {
if (!(attr.type == RKNN_TENSOR_UINT8 || attr.type == RKNN_TENSOR_INT8)) {
MMDEPLOY_ERROR("MMDeploy SDK only supports RKNN-INT8 model");
return Status(eInvalidArgument);
}
Expand Down
4 changes: 2 additions & 2 deletions docs/en/01-how-to-build/rockchip.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ It is recommended to create a virtual environment for the project.
2. Install RKNN python package following [rknn-toolkit2 doc](https://github.com/rockchip-linux/rknn-toolkit2/tree/master/doc) or [rknn-toolkit doc](https://github.com/rockchip-linux/rknn-toolkit/tree/master/doc). When installing rknn python package, it is better to append `--no-deps` after the commands to avoid dependency conflicts. RKNN-Toolkit2 package for example:

```
pip install packages/rknn_toolkit2-1.2.0_f7bb160f-cp36-cp36m-linux_x86_64.whl --no-deps
pip install packages/rknn_toolkit2-1.4.0_22dcfef4-cp36-cp36m-linux_x86_64.whl --no-deps
```

3. Install ONNX==1.8.0 before reinstall MMDeploy from source following the [instructions](../01-how-to-build/build_from_source.md). Note that there are conflicts between the pip dependencies of MMDeploy and RKNN. Here is the suggested packages versions for python 3.6:
Expand All @@ -50,7 +50,7 @@ Example:

```bash
python tools/deploy.py \
configs/mmpretrain/classification_rknn_static.py \
configs/mmpretrain/classification_rknn-fp16_static-224x224.py \
/mmpretrain_dir/configs/resnet/resnet50_8xb32_in1k.py \
https://download.openmmlab.com/mmclassification/v0/resnet/resnet50_batch256_imagenet_20200708-cfb998bf.pth \
/mmpretrain_dir/demo/demo.JPEG \
Expand Down
15 changes: 10 additions & 5 deletions docs/zh_cn/01-how-to-build/rockchip.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ MMDeploy 支持把模型部署到瑞芯微设备上。已支持的芯片：RV112
2.2 参考表中的安装指南，安装 RKNN python 安装包。建议在安装时，使用选项 `--no-deps`，以避免依赖包的冲突。以 rknn-toolkit2 为例:

```
pip install packages/rknn_toolkit2-1.2.0_f7bb160f-cp36-cp36m-linux_x86_64.whl --no-deps
pip install packages/rknn_toolkit2-1.4.0_22dcfef4-cp36-cp36m-linux_x86_64.whl --no-deps
```

2.3 先安装onnx==1.8.0,跟着 [instructions](../01-how-to-build/build_from_source.md)，源码安装 MMDeploy。 需要注意的是， MMDeploy 和 RKNN 依赖的安装包间有冲突的内容. 这里提供建议在 python 3.6 环境中使用的安装包版本:
Expand All @@ -88,7 +88,7 @@ git clone https://github.com/open-mmlab/mmpretrain
# 执行转换命令
cd /the/path/of/mmdeploy
python tools/deploy.py \
configs/mmpretrain/classification_rknn_static.py \
configs/mmpretrain/classification_rknn-fp16_static-224x224.py \
/the/path/of/mmpretrain/configs/resnet/resnet50_8xb32_in1k.py \
https://download.openmmlab.com/mmclassification/v0/resnet/resnet50_batch256_imagenet_20200708-cfb998bf.pth \
/the/path/of/mmpretrain/demo/demo.JPEG \
Expand Down Expand Up @@ -130,8 +130,13 @@ git clone https://github.com/open-mmlab/mmdetection

# 执行转换命令
python tools/deploy.py \
configs/mmpretrain/detection_rknn_static.py \

configs/mmdet/detection/detection_rknn-int8_static-320x320.py \
/the/path/of/mmdet/configs/yolov3/yolov3_mobilenetv2_320_300e_coco.py \
https://download.openmmlab.com/mmdetection/v2.0/yolo/yolov3_mobilenetv2_320_300e_coco/yolov3_mobilenetv2_320_300e_coco_20210719_215349-d18dff72.pth \
/the/path/of/mmdet/demo/demo.jpg \
--work-dir mmdeploy_models/mmdet/yolov3 \
--device cpu \
--dump-info
```

- RTMDet
Expand Down Expand Up @@ -299,7 +304,7 @@ export OpenCV_AARCH64_INSTALL_DIR=$(pwd)/install
cd /path/to/mmdeploy
mkdir -p build && cd build
export LD_LIBRARY_PATH=$RKNN_TOOL_CHAIN/lib64:$LD_LIBRARY_PATH
cmake \
cmake ..\
-DCMAKE_TOOLCHAIN_FILE=../cmake/toolchains/rknpu2-linux-gnu.cmake \
-DMMDEPLOY_BUILD_SDK=ON \
-DMMDEPLOY_TARGET_BACKENDS="rknn" \
Expand Down
9 changes: 8 additions & 1 deletion mmdeploy/backend/rknn/onnx2rknn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
from typing import Optional, Union

import mmengine
from packaging import version
from rknn.api import RKNN

from mmdeploy.utils import (get_common_config, get_normalization,
Expand Down Expand Up @@ -50,10 +51,16 @@ def onnx2rknn(onnx_model: str,
onnx_params = get_onnx_config(deploy_cfg)
quantization_cfg = get_quantization_config(deploy_cfg)

package_info = rknn_package_info()
input_names = onnx_params.get('input_names', None)
output_names = onnx_params.get('output_names', None)
input_size_list = get_backend_config(deploy_cfg).get(
'input_size_list', None)
# rknn-toolkit 1.5+ can not pass input output info, which is weird
if package_info['name'] == 'rknn-toolkit2' and version.parse(
package_info['version']) > version.parse('1.4'):
input_names, output_names, input_size_list = [None] * 3

# update norm value
if get_rknn_quantization(deploy_cfg) is True and model_cfg is not None:
transform = get_normalization(model_cfg)
Expand Down Expand Up @@ -85,7 +92,7 @@ def onnx2rknn(onnx_model: str,
if dataset_file is None:
quantization_cfg.update(dict(do_quantization=False))
logger.warning('no dataset passed in, quantization is skipped')
if rknn_package_info()['name'] == 'rknn-toolkit2':
if package_info['name'] == 'rknn-toolkit2':
quantization_cfg.pop('pre_compile', None)
ret = rknn.build(**quantization_cfg)
if ret != 0:
Expand Down