[Feat]: Support simcc from mmpose (#1187)

* add rewriting for simcc * add simccdecode for sdk * remove debug lines * fix cpp lint * move simcc decode to sdk * add simcc sdk config * update docs and regression yaml * update yml
2022-10-27 11:42:50 +08:00 · 2022-10-27 11:42:50 +08:00 · 0efc9e3c6d
parent c4285aed4a
commit 0efc9e3c6d
17 changed files with 317 additions and 54 deletions
--- a/configs/mmpose/pose-detection_simcc_ncnn_static-256x192.py
+++ b/configs/mmpose/pose-detection_simcc_ncnn_static-256x192.py
@ -0,0 +1,3 @@
+_base_ = ['./pose-detection_static.py', '../_base_/backends/ncnn.py']
+
+onnx_config = dict(input_shape=[192, 256], output_names=['simcc_x', 'simcc_y'])
--- a/configs/mmpose/pose-detection_simcc_onnxruntime_dynamic.py
+++ b/configs/mmpose/pose-detection_simcc_onnxruntime_dynamic.py
@ -0,0 +1,13 @@
+_base_ = ['./pose-detection_static.py', '../_base_/backends/onnxruntime.py']
+
+onnx_config = dict(
+    input_shape=[192, 256],
+    output_names=['simcc_x', 'simcc_y'],
+    dynamic_axes={
+        'input': {
+            0: 'batch',
+        },
+        'output': {
+            0: 'batch'
+        }
+    })
--- a/configs/mmpose/pose-detection_simcc_sdk_static-256x192.py
+++ b/configs/mmpose/pose-detection_simcc_sdk_static-256x192.py
@ -0,0 +1,12 @@
+_base_ = ['./pose-detection_static.py', '../_base_/backends/sdk.py']
+
+codebase_config = dict(model_type='sdk')
+onnx_config = dict(output_names=['simcc_x', 'simcc_y'])
+
+backend_config = dict(pipeline=[
+    dict(type='LoadImageFromFile'),
+    dict(type='GetBBoxCenterScale'),
+    dict(type='PackPoseInputs')
+])
+
+ext_info = dict(image_size=[192, 256], padding=1.25)
--- a/configs/mmpose/pose-detection_simcc_tensorrt_dynamic-256x192.py
+++ b/configs/mmpose/pose-detection_simcc_tensorrt_dynamic-256x192.py
@ -0,0 +1,24 @@
+_base_ = ['./pose-detection_static.py', '../_base_/backends/tensorrt.py']
+
+onnx_config = dict(
+    input_shape=[192, 256],
+    output_names=['simcc_x', 'simcc_y'],
+    dynamic_axes={
+        'input': {
+            0: 'batch',
+        },
+        'output': {
+            0: 'batch'
+        }
+    })
+
+backend_config = dict(
+    common_config=dict(max_workspace_size=1 << 30),
+    model_inputs=[
+        dict(
+            input_shapes=dict(
+                input=dict(
+                    min_shape=[1, 3, 256, 192],
+                    opt_shape=[2, 3, 256, 192],
+                    max_shape=[4, 3, 256, 192])))
+    ])
--- a/csrc/mmdeploy/codebase/mmpose/simcc_label.cpp
+++ b/csrc/mmdeploy/codebase/mmpose/simcc_label.cpp
@ -0,0 +1,123 @@
+// Copyright (c) OpenMMLab. All rights reserved.
+
+#include <cctype>
+#include <opencv2/imgproc.hpp>
+
+#include "mmdeploy/core/device.h"
+#include "mmdeploy/core/registry.h"
+#include "mmdeploy/core/serialization.h"
+#include "mmdeploy/core/tensor.h"
+#include "mmdeploy/core/utils/device_utils.h"
+#include "mmdeploy/core/utils/formatter.h"
+#include "mmdeploy/core/value.h"
+#include "mmdeploy/experimental/module_adapter.h"
+#include "mmpose.h"
+#include "opencv_utils.h"
+
+namespace mmdeploy::mmpose {
+
+using std::string;
+using std::vector;
+
+template <class F>
+struct _LoopBody : public cv::ParallelLoopBody {
+  F f_;
+  _LoopBody(F f) : f_(std::move(f)) {}
+  void operator()(const cv::Range& range) const override { f_(range); }
+};
+
+class SimCCLabelDecode : public MMPose {
+ public:
+  explicit SimCCLabelDecode(const Value& config) : MMPose(config) {
+    if (config.contains("params")) {
+      auto& params = config["params"];
+      flip_test_ = params.value("flip_test", flip_test_);
+      simcc_split_ratio_ = params.value("simcc_split_ratio", simcc_split_ratio_);
+      if (params.contains("input_size")) {
+        from_value(params["input_size"], input_size_);
+      }
+    }
+  }
+
+  Result<Value> operator()(const Value& _data, const Value& _prob) {
+    MMDEPLOY_DEBUG("preprocess_result: {}", _data);
+    MMDEPLOY_DEBUG("inference_result: {}", _prob);
+
+    Device cpu_device{"cpu"};
+    OUTCOME_TRY(auto simcc_x,
+                MakeAvailableOnDevice(_prob["simcc_x"].get<Tensor>(), cpu_device, stream()));
+    OUTCOME_TRY(auto simcc_y,
+                MakeAvailableOnDevice(_prob["simcc_y"].get<Tensor>(), cpu_device, stream()));
+    OUTCOME_TRY(stream().Wait());
+    if (!(simcc_x.shape().size() == 3 && simcc_x.data_type() == DataType::kFLOAT)) {
+      MMDEPLOY_ERROR("unsupported `simcc_x` tensor, shape: {}, dtype: {}", simcc_x.shape(),
+                     (int)simcc_x.data_type());
+      return Status(eNotSupported);
+    }
+
+    auto& img_metas = _data["img_metas"];
+
+    Tensor keypoints({Device{"cpu"}, DataType::kFLOAT, {simcc_x.shape(0), simcc_x.shape(1), 2}});
+    Tensor scores({Device{"cpu"}, DataType::kFLOAT, {simcc_x.shape(0), simcc_x.shape(1), 1}});
+    get_simcc_maximum(simcc_x, simcc_y, keypoints, scores);
+
+    std::vector<float> center;
+    std::vector<float> scale;
+    from_value(img_metas["center"], center);
+    from_value(img_metas["scale"], scale);
+    PoseDetectorOutput output;
+
+    float* keypoints_data = keypoints.data<float>();
+    float* scores_data = scores.data<float>();
+    float scale_value = 200, x = -1, y = -1, s = 0;
+    for (int i = 0; i < simcc_x.shape(1); i++) {
+      x = *(keypoints_data + 0) / simcc_split_ratio_;
+      y = *(keypoints_data + 1) / simcc_split_ratio_;
+      x = x * scale[0] * scale_value / input_size_[0] + center[0] - scale[0] * scale_value * 0.5;
+      y = y * scale[1] * scale_value / input_size_[1] + center[1] - scale[1] * scale_value * 0.5;
+      s = *(scores_data + 0);
+      output.key_points.push_back({{x, y}, s});
+      keypoints_data += 2;
+      scores_data += 1;
+    }
+    return to_value(output);
+  }
+
+  void get_simcc_maximum(const Tensor& simcc_x, const Tensor& simcc_y, Tensor& keypoints,
+                         Tensor& scores) {
+    int K = simcc_x.shape(1);
+    int N_x = simcc_x.shape(2);
+    int N_y = simcc_y.shape(2);
+
+    cv::parallel_for_(cv::Range(0, K), _LoopBody{[&](const cv::Range& r) {
+                        for (int i = r.start; i < r.end; i++) {
+                          float* data_x = const_cast<float*>(simcc_x.data<float>()) + i * N_x;
+                          float* data_y = const_cast<float*>(simcc_y.data<float>()) + i * N_y;
+                          cv::Mat mat_x = cv::Mat(N_x, 1, CV_32FC1, data_x);
+                          cv::Mat mat_y = cv::Mat(N_y, 1, CV_32FC1, data_y);
+                          double min_val_x, max_val_x, min_val_y, max_val_y;
+                          cv::Point min_loc_x, max_loc_x, min_loc_y, max_loc_y;
+                          cv::minMaxLoc(mat_x, &min_val_x, &max_val_x, &min_loc_x, &max_loc_x);
+                          cv::minMaxLoc(mat_y, &min_val_y, &max_val_y, &min_loc_y, &max_loc_y);
+                          float s = max_val_x > max_val_y ? max_val_y : max_val_x;
+                          float x = s > 0 ? max_loc_x.y : -1.0;
+                          float y = s > 0 ? max_loc_y.y : -1.0;
+                          float* keypoints_data = keypoints.data<float>() + i * 2;
+                          float* scores_data = scores.data<float>() + i;
+                          *(scores_data) = s;
+                          *(keypoints_data + 0) = x;
+                          *(keypoints_data + 1) = y;
+                        }
+                      }});
+  }
+
+ private:
+  bool flip_test_{false};
+  bool shift_heatmap_{false};
+  float simcc_split_ratio_{2.0};
+  std::vector<int> input_size_{192, 256};
+};
+
+REGISTER_CODEBASE_COMPONENT(MMPose, SimCCLabelDecode);
+
+}  // namespace mmdeploy::mmpose
--- a/docs/en/03-benchmark/benchmark.md
+++ b/docs/en/03-benchmark/benchmark.md
@ -1710,6 +1710,27 @@ Users can directly test the performance through [how_to_evaluate_a_model.md](../
    <td align="center">-</td>
    <td align="center">0.774</td>
  </tr>
+  <tr>
+    <td align="center" rowspan="2"><a href="https://github.com/open-mmlab/mmpose/blob/1.x/configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py">SimCC</a></td>
+    <td align="center" rowspan="2">Pose Detection</td>
+    <td align="center" rowspan="2">COCO</td>
+    <td align="center">AP</td>
+    <td align="center">0.607</td>
+    <td align="center">-</td>
+    <td align="center">0.608</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+  </tr>
+  <tr>
+    <td align="center">AR</td>
+    <td align="center">0.668</td>
+    <td align="center">-</td>
+    <td align="center">0.672</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+  </tr>
 </tbody>
 </table>
 </div>
--- a/docs/en/03-benchmark/supported_models.md
+++ b/docs/en/03-benchmark/supported_models.md
@ -74,6 +74,7 @@ The table below lists the models that are guaranteed to be exportable to other b
 | MSPN                        | MMPose           |      N      |      Y      |    Y     |  Y   |   N   |    Y     |   N    |  N   |     [config](https://mmpose.readthedocs.io/en/latest/papers/backbones.html#mspn-arxiv-2019)     |
 | LiteHRNet                   | MMPose           |      N      |      Y      |    Y     |  N   |   N   |    Y     |   N    |  N   |   [config](https://mmpose.readthedocs.io/en/latest/papers/backbones.html#litehrnet-cvpr-2021)   |
 | Hourglass                   | MMPose           |      N      |      Y      |    Y     |  Y   |   N   |    Y     |   N    |  N   |   [config](https://mmpose.readthedocs.io/en/latest/papers/backbones.html#hourglass-eccv-2016)   |
+| SimCC                       | MMPose           |      N      |      Y      |    Y     |  Y   |   N   |    N     |   N    |  N   | [config](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/algorithms.html#simcc-eccv-2022) |
 | PointPillars                | MMDetection3d    |      ?      |      Y      |    Y     |  N   |   N   |    Y     |   N    |  N   |     [config](https://github.com/open-mmlab/mmdetection3d/blob/master/configs/pointpillars)      |
 | CenterPoint (pillar)        | MMDetection3d    |      ?      |      Y      |    Y     |  N   |   N   |    Y     |   N    |  N   |      [config](https://github.com/open-mmlab/mmdetection3d/blob/master/configs/centerpoint)      |
 | RotatedRetinaNet            | RotatedDetection |      N      |      Y      |    Y     |  N   |   N   |    N     |   N    |  N   | [config](https://github.com/open-mmlab/mmrotate/blob/main/configs/rotated_retinanet/README.md)  |
--- a/docs/en/04-supported-codebases/mmpose.md
+++ b/docs/en/04-supported-codebases/mmpose.md
@ -1,14 +1,15 @@
 # MMPose Deployment

- [Installation](#installation)
-  - [Install mmpose](#install-mmpose)
-  - [Install mmdeploy](#install-mmdeploy)
- [Convert model](#convert-model)
- [Model specification](#model-specification)
- [Model inference](#model-inference)
-  - [Backend model inference](#backend-model-inference)
-  - [SDK model inference](#sdk-model-inference)
- [Supported models](#supported-models)
+- [MMPose Deployment](#mmpose-deployment)
+  - [Installation](#installation)
+    - [Install mmpose](#install-mmpose)
+    - [Install mmdeploy](#install-mmdeploy)
+  - [Convert model](#convert-model)
+  - [Model specification](#model-specification)
+  - [Model inference](#model-inference)
+    - [Backend model inference](#backend-model-inference)
+    - [SDK model inference](#sdk-model-inference)
+  - [Supported models](#supported-models)

 ______________________________________________________________________

@ -151,8 +152,10 @@ TODO

 ## Supported models

-| Model                                                                                                 | Task          | ONNX Runtime | TensorRT | ncnn | PPLNN | OpenVINO |
-| :---------------------------------------------------------------------------------------------------- | :------------ | :----------: | :------: | :--: | :---: | :------: |
-| [HRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#hrnet-cvpr-2019)         | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
-| [MSPN](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#mspn-arxiv-2019)          | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
-| [LiteHRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#litehrnet-cvpr-2021) | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| Model                                                                                                  | Task          | ONNX Runtime | TensorRT | ncnn | PPLNN | OpenVINO |
+| :----------------------------------------------------------------------------------------------------- | :------------ | :----------: | :------: | :--: | :---: | :------: |
+| [HRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#hrnet-cvpr-2019)          | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [MSPN](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#mspn-arxiv-2019)           | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [LiteHRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#litehrnet-cvpr-2021)  | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [Hourglass](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/algorithms.html#hourglass-eccv-2016) | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [SimCC](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/algorithms.html#simcc-eccv-2022)         | PoseDetection |      Y       |    Y     |  Y   |   N   |    N     |
--- a/docs/zh_cn/03-benchmark/benchmark.md
+++ b/docs/zh_cn/03-benchmark/benchmark.md
@ -1686,6 +1686,27 @@ GPU: ncnn, TensorRT, PPLNN
    <td align="center">-</td>
    <td align="center">0.774</td>
  </tr>
+  <tr>
+    <td align="center" rowspan="2"><a href="https://github.com/open-mmlab/mmpose/blob/1.x/configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py">SimCC</a></td>
+    <td align="center" rowspan="2">Pose Detection</td>
+    <td align="center" rowspan="2">COCO</td>
+    <td align="center">AP</td>
+    <td align="center">0.607</td>
+    <td align="center">-</td>
+    <td align="center">0.608</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+  </tr>
+  <tr>
+    <td align="center">AR</td>
+    <td align="center">0.668</td>
+    <td align="center">-</td>
+    <td align="center">0.672</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+    <td align="center">-</td>
+  </tr>
 </tbody>
 </table>
 </div>
--- a/docs/zh_cn/03-benchmark/supported_models.md
+++ b/docs/zh_cn/03-benchmark/supported_models.md
@ -74,6 +74,7 @@
 | MSPN                        | MMPose           |      N      |      Y      |    Y     |  Y   |   N   |    Y     |   N    |  N   |     [config](https://mmpose.readthedocs.io/en/latest/papers/backbones.html#mspn-arxiv-2019)     |
 | LiteHRNet                   | MMPose           |      N      |      Y      |    Y     |  N   |   N   |    Y     |   N    |  N   |   [config](https://mmpose.readthedocs.io/en/latest/papers/backbones.html#litehrnet-cvpr-2021)   |
 | Hourglass                   | MMPose           |      N      |      Y      |    Y     |  Y   |   N   |    Y     |   N    |  N   |   [config](https://mmpose.readthedocs.io/en/latest/papers/backbones.html#hourglass-eccv-2016)   |
+| SimCC                       | MMPose           |      N      |      Y      |    Y     |  Y   |   N   |    N     |   N    |  N   | [config](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/algorithms.html#simcc-eccv-2022) |
 | PointPillars                | MMDetection3d    |      ?      |      Y      |    Y     |  N   |   N   |    Y     |   N    |  N   |     [config](https://github.com/open-mmlab/mmdetection3d/blob/master/configs/pointpillars)      |
 | CenterPoint (pillar)        | MMDetection3d    |      ?      |      Y      |    Y     |  N   |   N   |    Y     |   N    |  N   |      [config](https://github.com/open-mmlab/mmdetection3d/blob/master/configs/centerpoint)      |
 | RotatedRetinaNet            | RotatedDetection |      N      |      Y      |    Y     |  N   |   N   |    N     |   N    |  N   | [config](https://github.com/open-mmlab/mmrotate/blob/main/configs/rotated_retinanet/README.md)  |
--- a/docs/zh_cn/04-supported-codebases/mmpose.md
+++ b/docs/zh_cn/04-supported-codebases/mmpose.md
@ -1,14 +1,15 @@
 # MMPose 模型部署

- [安装](#安装)
-  - [安装 mmcls](#安装-mmpose)
-  - [安装 mmdeploy](#安装-mmdeploy)
- [模型转换](#模型转换)
- [模型规范](#模型规范)
- [模型推理](#模型推理)
-  - [后端模型推理](#后端模型推理)
-  - [SDK 模型推理](#sdk-模型推理)
- [模型支持列表](#模型支持列表)
+- [MMPose 模型部署](#mmpose-模型部署)
+  - [安装](#安装)
+    - [安装 mmpose](#安装-mmpose)
+    - [安装 mmdeploy](#安装-mmdeploy)
+  - [模型转换](#模型转换)
+  - [模型规范](#模型规范)
+  - [模型推理](#模型推理)
+    - [后端模型推理](#后端模型推理)
+    - [SDK 模型推理](#sdk-模型推理)
+  - [模型支持列表](#模型支持列表)

 ______________________________________________________________________

@ -155,8 +156,10 @@ task_processor.visualize(

 ## 模型支持列表

-| Model                                                                                                 | Task          | ONNX Runtime | TensorRT | ncnn | PPLNN | OpenVINO |
-| :---------------------------------------------------------------------------------------------------- | :------------ | :----------: | :------: | :--: | :---: | :------: |
-| [HRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#hrnet-cvpr-2019)         | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
-| [MSPN](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#mspn-arxiv-2019)          | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
-| [LiteHRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#litehrnet-cvpr-2021) | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| Model                                                                                                  | Task          | ONNX Runtime | TensorRT | ncnn | PPLNN | OpenVINO |
+| :----------------------------------------------------------------------------------------------------- | :------------ | :----------: | :------: | :--: | :---: | :------: |
+| [HRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#hrnet-cvpr-2019)          | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [MSPN](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#mspn-arxiv-2019)           | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [LiteHRNet](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/backbones.html#litehrnet-cvpr-2021)  | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [Hourglass](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/algorithms.html#hourglass-eccv-2016) | PoseDetection |      Y       |    Y     |  Y   |   N   |    Y     |
+| [SimCC](https://mmpose.readthedocs.io/en/1.x/model_zoo_papers/algorithms.html#simcc-eccv-2022)         | PoseDetection |      Y       |    Y     |  Y   |   N   |    N     |
--- a/mmdeploy/codebase/mmpose/deploy/pose_detection.py
+++ b/mmdeploy/codebase/mmpose/deploy/pose_detection.py
@ -333,18 +333,25 @@ class PoseDetection(BaseTask):

    def get_postprocess(self, *args, **kwargs) -> Dict:
        """Get the postprocess information for SDK."""
+        codec = self.model_cfg.codec
+        if isinstance(codec, (list, tuple)):
+            codec = codec[-1]
        component = 'UNKNOWN'
        params = copy.deepcopy(self.model_cfg.model.test_cfg)
+        params.update(codec)
        if self.model_cfg.model.type == 'TopdownPoseEstimator':
-            head_type = self.model_cfg.model.head.type
-            if head_type == 'HeatmapHead':
+            component = 'TopdownHeatmapSimpleHeadDecode'
+            if codec.type == 'MSRAHeatmap':
                params['post_process'] = 'default'
-                component = 'TopdownHeatmapSimpleHeadDecode'
-            elif head_type == 'MSPNHead':
+            elif codec.type == 'UDPHeatmap':
+                params['post_process'] = 'default'
+                params['use_udp'] = True
+            elif codec.type == 'MegviiHeatmap':
                params['post_process'] = 'megvii'
                params['modulate_kernel'] = self.model_cfg.kernel_sizes[-1]
-                component = 'TopdownHeatmapMSMUHeadDecode'
+            elif codec.type == 'SimCCLabel':
+                component = 'SimCCLabelDecode'
            else:
-                raise RuntimeError(f'Unsupported head type: {head_type}')
+                raise RuntimeError(f'Unsupported codecs type: {codec.type}')
        postprocess = dict(params=params, type=component)
        return postprocess
--- a/mmdeploy/codebase/mmpose/deploy/pose_detection_model.py
+++ b/mmdeploy/codebase/mmpose/deploy/pose_detection_model.py
@ -101,24 +101,14 @@ class End2EndModel(BaseBackendModel):
        inputs = inputs.contiguous().to(self.device)
        batch_outputs = self.wrapper({self.input_name: inputs})
        batch_outputs = self.wrapper.output_to_list(batch_outputs)
-        batch_heatmaps = batch_outputs[0]
-        # flip test
-        test_cfg = self.model_cfg.model.test_cfg
-        if test_cfg.get('flip_test', False):
-            from mmpose.models.utils.tta import flip_heatmaps
-            batch_inputs_flip = inputs.flip(-1).contiguous()
-            batch_outputs_flip = self.wrapper(
-                {self.input_name: batch_inputs_flip})
-            batch_heatmaps_flip = self.wrapper.output_to_list(
-                batch_outputs_flip)[0]
-            flip_indices = data_samples[0].metainfo['flip_indices']
-            batch_heatmaps_flip = flip_heatmaps(
-                batch_heatmaps_flip,
-                flip_mode=test_cfg.get('flip_mode', 'heatmap'),
-                flip_indices=flip_indices,
-                shift_heatmap=test_cfg.get('shift_heatmap', False))
-            batch_heatmaps = (batch_heatmaps + batch_heatmaps_flip) * 0.5
-        preds = self.head.decode(batch_heatmaps)
+        codec = self.model_cfg.codec
+        if isinstance(codec, (list, tuple)):
+            codec = codec[-1]
+        if codec.type == 'SimCCLabel':
+            batch_pred_x, batch_pred_y = batch_outputs
+            preds = self.head.decode((batch_pred_x, batch_pred_y))
+        else:
+            preds = self.head.decode(batch_outputs[0])
        results = self.pack_result(preds, data_samples)
        return results

--- a/mmdeploy/codebase/mmpose/models/heads/init.py
+++ b/mmdeploy/codebase/mmpose/models/heads/init.py
@ -1,4 +1,4 @@
 # Copyright (c) OpenMMLab. All rights reserved.
-from . import heatmap_head, mspn_head, regression_head
+from . import heatmap_head, mspn_head, regression_head, simcc_head

-__all__ = ['heatmap_head', 'mspn_head', 'regression_head']
+__all__ = ['heatmap_head', 'mspn_head', 'regression_head', 'simcc_head']
--- a/mmdeploy/codebase/mmpose/models/heads/simcc_head.py
+++ b/mmdeploy/codebase/mmpose/models/heads/simcc_head.py
@ -0,0 +1,22 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmdeploy.core import FUNCTION_REWRITER
+
+
+@FUNCTION_REWRITER.register_rewriter(
+    'mmpose.models.heads.heatmap_heads.SimCCHead.predict')
+def simcc_head__predict(ctx, self, feats, batch_data_samples, test_cfg=None):
+    """Rewrite `predict` of HeatmapHead for default backend.
+
+    1. skip decoding and return output tensor directly.
+
+    Args:
+        feats (tuple[Tensor]): Input features.
+        batch_data_samples (list[SampleList]): Data samples contain
+            image meta information.
+        test_cfg (ConfigType): test config.
+
+    Returns:
+        output_heatmap (torch.Tensor): Output heatmaps.
+    """
+    simcc_x, simcc_y = self.forward(feats)
+    return simcc_x, simcc_y
--- a/mmdeploy/pytorch/functions/flatten.py
+++ b/mmdeploy/pytorch/functions/flatten.py
@ -5,6 +5,10 @@ from mmdeploy.core import FUNCTION_REWRITER
 from mmdeploy.utils import Backend


+@FUNCTION_REWRITER.register_rewriter(
+    func_name='torch.Tensor.flatten', backend=Backend.NCNN.value)
+@FUNCTION_REWRITER.register_rewriter(
+    func_name='torch.flatten', backend=Backend.NCNN.value)
@FUNCTION_REWRITER.register_rewriter(
    func_name='torch.Tensor.flatten', backend=Backend.COREML.value)
@FUNCTION_REWRITER.register_rewriter(
--- a/tests/regression/mmpose.yml
+++ b/tests/regression/mmpose.yml
@ -124,3 +124,18 @@ models:
 #      - *pipeline_trt_static_fp32_256x256
 #      - *pipeline_ncnn_static_fp32_256x256
 #      - *pipeline_openvino_static_fp32_256x256
+
+# TODO: no mobilenetv2_coco.yml in latest mmpose, enable this later
+#  - name: SimCC
+#    metafile: configs/body_2d_keypoint/simcc/coco/mobilenetv2_coco.yml
+#    model_configs:
+#      - configs/body_2d_keypoint/simcc/coco/simcc_mobilenetv2_wo-deconv-8xb64-210e_coco-256x192.py
+#    pipelines:
+#      - convert_image: *convert_image
+#        deploy_config: configs/mmpose/pose-detection_simcc_onnxruntime_dynamic.py
+#      - convert_image: *convert_image
+#        deploy_config: configs/mmpose/pose-detection_simcc_tensorrt_dynamic-256x192.py
+#        backend_test: *default_backend_test
+#        sdk_config: configs/mmpose/pose-detection_simcc_sdk_static-256x192.py
+#      - convert_image: *convert_image
+#        deploy_config: configs/mmpose/pose-detection_simcc_ncnn_static-256x192.py