[Quantization] Add new PaddleSeg models quantization support. (#953)

* Fix links in readme * Fix links in readme * Update PPOCRv2/v3 examples * Update auto compression configs * Add neww quantization support for paddleclas model * Update quantized Yolov6s model download link * Improve PPOCR comments * Add English doc for quantization * Fix PPOCR rec model bug * Add new paddleseg quantization support * Add new paddleseg quantization support * Add new paddleseg quantization support * Add new paddleseg quantization support
2025-10-08 10:00:29 +08:00 · 2022-12-25 16:33:57 +08:00
parent 2431890f73
commit 2d3d941372
5 changed files with 120 additions and 3 deletions
--- a/tools/common_tools/auto_compression/README.md
+++ b/tools/common_tools/auto_compression/README.md
@@ -103,13 +103,15 @@ FastDeploy目前为用户提供了多个模型的压缩[config](./configs/)文
 | [mobilenetv3_large_x1_0_quant](./configs/classification/mobilenetv3_large_x1_0_quant.yaml)      |   [mobilenetv3_large_x1_0](https://bj.bcebos.com/paddlehub/fastdeploy/MobileNetV3_large_x1_0_ssld_infer.tgz)          |     |
 | [pphgnet_tiny_quant](./configs/classification/pphgnet_tiny_quant.yaml)      |   [pphgnet_tiny](https://bj.bcebos.com/paddlehub/fastdeploy/PPHGNet_tiny_ssld_infer.tgz)          |     |
 | [pplcnetv2_base_quant](./configs/classification/pplcnetv2_base_quant.yaml)      |   [pplcnetv2_base](https://bj.bcebos.com/paddlehub/fastdeploy/PPLCNetV2_base_infer.tgz)          |     |
-
 | [yolov5s_quant](./configs/detection/yolov5s_quant.yaml)       |   [yolov5s](https://paddle-slim-models.bj.bcebos.com/act/yolov5s.onnx)         |     |
 | [yolov6s_quant](./configs/detection/yolov6s_quant.yaml)       |  [yolov6s](https://paddle-slim-models.bj.bcebos.com/act/yolov6s.onnx)          |     |
 | [yolov7_quant](./configs/detection/yolov7_quant.yaml)        | [yolov7](https://paddle-slim-models.bj.bcebos.com/act/yolov7.onnx)           |      |
 | [ppyoloe_withNMS_quant](./configs/detection/ppyoloe_withNMS_quant.yaml)       |  [ppyoloe_l](https://bj.bcebos.com/v1/paddle-slim-models/act/ppyoloe_crn_l_300e_coco.tar)    | 支持PPYOLOE的s,m,l,x系列模型, 从PaddleDetection导出模型时正常导出, 不要去除NMS |
 | [ppyoloe_plus_withNMS_quant](./configs/detection/ppyoloe_plus_withNMS_quant.yaml)       |  [ppyoloe_plus_s](https://bj.bcebos.com/paddlehub/fastdeploy/ppyoloe_plus_crn_s_80e_coco.tar)    | 支持PPYOLOE+的s,m,l,x系列模型, 从PaddleDetection导出模型时正常导出, 不要去除NMS |
-| [pp_liteseg_quant](./configs/segmentation/pp_liteseg_quant.yaml)    |   [pp_liteseg](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer.tgz)        |       |
+| [pp_liteseg_quant](./configs/segmentation/pp_liteseg_quant.yaml)    |   [pp_liteseg](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer.tgz)        |
+| [deeplabv3_resnet_quant](./configs/segmentation/deeplabv3_resnet_quant.yaml)    |   [deeplabv3_resnet101](https://bj.bcebos.com/paddlehub/fastdeploy/Deeplabv3_ResNet101_OS8_cityscapes_without_argmax_infer.tgz)        |       |
+| [fcn_hrnet_quant](./configs/segmentation/fcn_hrnet_quant.yaml)    |   [fcn_hrnet](https://bj.bcebos.com/paddlehub/fastdeploy/FCN_HRNet_W18_cityscapes_without_argmax_infer.tgz)        |       |
+| [unet_quant](./configs/segmentation/unet_quant.yaml)    |   [unet](https://bj.bcebos.com/paddlehub/fastdeploy/Unet_cityscapes_without_argmax_infer.tgz)        |       |      |



--- a/tools/common_tools/auto_compression/README_EN.md
+++ b/tools/common_tools/auto_compression/README_EN.md
@@ -103,7 +103,7 @@ To complete the quantization, developers only need to provide a customized model
 ## 3. FastDeploy One-Click Model Auto Compression Config file examples
 FastDeploy currently provides users with compression [config](./configs/) files of multiple models, and the corresponding FP32 model, Users can directly download and experience it.

-| Config文件                | 待压缩的FP32模型 | 备注                                                       |
+| Config file                | FP32 model | Note                                                       |
 | -------------------- | ------------------------------------------------------------ |----------------------------------------- |
 | [mobilenetv1_ssld_quant](./configs/classification/mobilenetv1_ssld_quant.yaml)      | [mobilenetv1_ssld](https://bj.bcebos.com/paddlehub/fastdeploy/MobileNetV1_ssld_infer.tgz)           |           |
 | [resnet50_vd_quant](./configs/classification/resnet50_vd_quant.yaml)      |   [resnet50_vd](https://bj.bcebos.com/paddlehub/fastdeploy/ResNet50_vd_infer.tgz)          |     |
@@ -117,6 +117,10 @@ FastDeploy currently provides users with compression [config](./configs/) files
 | [ppyoloe_withNMS_quant](./configs/detection/ppyoloe_withNMS_quant.yaml)       |  [ppyoloe_l](https://bj.bcebos.com/v1/paddle-slim-models/act/ppyoloe_crn_l_300e_coco.tar)    | Support PPYOLOE's s,m,l,x series models, export the model normally when exporting the model from PaddleDetection, do not remove NMS |
 | [ppyoloe_plus_withNMS_quant](./configs/detection/ppyoloe_plus_withNMS_quant.yaml)       |  [ppyoloe_plus_s](https://bj.bcebos.com/paddlehub/fastdeploy/ppyoloe_plus_crn_s_80e_coco.tar)    | Support PPYOLOE+'s s,m,l,x series models, export the model normally when exporting the model from PaddleDetection, do not remove NMS |
 | [pp_liteseg_quant](./configs/segmentation/pp_liteseg_quant.yaml)    |   [pp_liteseg](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer.tgz)        |       |
+| [deeplabv3_resnet_quant](./configs/segmentation/deeplabv3_resnet_quant.yaml)    |   [deeplabv3_resnet101](https://bj.bcebos.com/paddlehub/fastdeploy/Deeplabv3_ResNet101_OS8_cityscapes_without_argmax_infer.tgz)        |       |
+| [fcn_hrnet_quant](./configs/segmentation/fcn_hrnet_quant.yaml)    |   [fcn_hrnet](https://bj.bcebos.com/paddlehub/fastdeploy/FCN_HRNet_W18_cityscapes_without_argmax_infer.tgz)        |       |
+| [unet_quant](./configs/segmentation/unet_quant.yaml)    |   [unet](https://bj.bcebos.com/paddlehub/fastdeploy/Unet_cityscapes_without_argmax_infer.tgz)        |       |      |
+

 ## 3. Deploy quantized models on FastDeploy

--- a/tools/common_tools/auto_compression/configs/segmentation/deeplabv3_resnet_quant.yaml
+++ b/tools/common_tools/auto_compression/configs/segmentation/deeplabv3_resnet_quant.yaml
@@ -0,0 +1,37 @@
+Global:
+  model_dir: ./Deeplabv3_ResNet101_OS8_cityscapes_without_argmax_infer/
+  format: paddle
+  model_filename: model.pdmodel
+  params_filename: model.pdiparams
+  qat_image_path: ./train_stuttgart
+  ptq_image_path: ./train_stuttgart
+  input_list: ['x']
+  qat_preprocess: ppseg_cityscapes_qat_preprocess
+  ptq_preprocess: ppseg_cityscapes_ptq_preprocess
+  qat_batch_size: 2
+
+
+Distillation:
+  alpha: 1.0
+  loss: l2
+  node:
+  - conv2d_225.tmp_0
+
+QuantAware:
+  onnx_format: True
+  quantize_op_types:
+  - conv2d
+  - depthwise_conv2d
+
+TrainConfig:
+  epochs: 1
+  eval_iter: 360
+  learning_rate: 0.0001
+  optimizer_builder:
+    optimizer:
+      type: SGD
+    weight_decay: 0.0005
+
+PTQ:
+  calibration_method: 'avg'   # option: avg, abs_max, hist, KL, mse
+  skip_tensor_list: None
--- a/tools/common_tools/auto_compression/configs/segmentation/fcn_hrnet_quant.yaml
+++ b/tools/common_tools/auto_compression/configs/segmentation/fcn_hrnet_quant.yaml
@@ -0,0 +1,36 @@
+Global:
+  model_dir: ./FCN_HRNet_W18_cityscapes_without_argmax_infer
+  format: paddle
+  model_filename: model.pdmodel
+  params_filename: model.pdiparams
+  qat_image_path: ./train_stuttgart
+  ptq_image_path: ./train_stuttgart
+  input_list: ['x']
+  qat_preprocess: ppseg_cityscapes_qat_preprocess
+  ptq_preprocess: ppseg_cityscapes_ptq_preprocess
+  qat_batch_size: 2
+
+Distillation:
+  alpha: 1.0
+  loss: l2
+  node:
+  - conv2d_613.tmp_1
+
+QuantAware:
+  onnx_format: True
+  quantize_op_types:
+  - conv2d
+  - depthwise_conv2d
+
+TrainConfig:
+  epochs: 20
+  eval_iter: 360
+  learning_rate: 0.0001
+  optimizer_builder:
+    optimizer:
+      type: SGD
+    weight_decay: 4.0e-05
+
+PTQ:
+  calibration_method: 'avg'   # option: avg, abs_max, hist, KL, mse
+  skip_tensor_list: None
--- a/tools/common_tools/auto_compression/configs/segmentation/unet_quant.yaml
+++ b/tools/common_tools/auto_compression/configs/segmentation/unet_quant.yaml
@@ -0,0 +1,38 @@
+Global:
+  model_dir: ./Unet_cityscapes_without_argmax_infer/
+  format: paddle
+  model_filename: model.pdmodel
+  params_filename: model.pdiparams
+  qat_image_path: ./train_stuttgart
+  ptq_image_path: ./train_stuttgart
+  input_list: ['x']
+  qat_preprocess: ppseg_cityscapes_qat_preprocess
+  ptq_preprocess: ppseg_cityscapes_ptq_preprocess
+  qat_batch_size: 2
+
+
+Distillation:
+  alpha: 1.0
+  loss: l2
+  node:
+  - conv2d_37.tmp_1
+
+QuantAware:
+  onnx_format: True
+  quantize_op_types:
+  - conv2d
+  - depthwise_conv2d
+
+
+TrainConfig:
+  epochs: 10
+  eval_iter: 180
+  learning_rate: 0.0005
+  optimizer_builder:
+    optimizer:
+      type: SGD
+    weight_decay: 4.0e-05
+
+PTQ:
+  calibration_method: 'avg'   # option: avg, abs_max, hist, KL, mse
+  skip_tensor_list: None