fix the config of ssd_mobilenet_v1 (PaddlePaddle#1443)

467718584 · Oct 8, 2022 · 253c09b · 253c09b
1 parent d2acb14
commit 253c09b
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 11 deletions.
diff --git a/example/auto_compression/detection/README.md b/example/auto_compression/detection/README.md
@@ -27,6 +27,11 @@
 - mAP的指标均在COCO val2017数据集中评测得到，IoU=0.5:0.95。
 - PP-YOLOE-l模型在Tesla V100的GPU环境下测试，并且开启TensorRT，batch_size=1，包含NMS，测试脚本是[benchmark demo](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/deploy/python)。
 - PP-YOLOE-s模型在Tesla T4，TensorRT 8.4.1，CUDA 11.2，batch_size=1，不包含NMS，测试脚本是[cpp_infer_ppyoloe](./cpp_infer_ppyoloe)。
+### SSD on Pascal VOC
+| 模型  | Box AP | ACT量化Box AP | TRT-FP32 | TRT-INT8 |  配置文件 | 量化模型  |
+| :-------- |:-------- | :---------------------: | :----------------: | :---------------: | :----------------------: | :---------------------: |
+| SSD-MobileNetv1 |  73.8  |   73.52    |   4.0ms  |  1.7ms  |  [config](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/example/auto_compression/detection/configs/ssd_mbv1_voc_qat_dis.yaml) | [Model](https://bj.bcebos.com/v1/paddle-slim-models/act/ssd_mobilenet_v1_quant.tar) |
+- 测速环境：Tesla T4，TensorRT 8.4.1，CUDA 11.2，batch_size=1，包含NMS.
 ## 3. 自动压缩流程
 
 #### 3.1 准备环境

diff --git a/example/auto_compression/detection/configs/ssd_mbv1_voc_qat_dis.yaml b/example/auto_compression/detection/configs/ssd_mbv1_voc_qat_dis.yaml
@@ -2,32 +2,33 @@ Global:
   reader_config: configs/ssd_reader.yml
   input_list: ['image', 'scale_factor', 'im_shape']
   Evaluation: True
-  model_dir: ./ssd_mobilenet_v1_300_120e_voc
+  model_dir: ./ssd_mobilenet_v1_300_120e_voc # Model Link: https://bj.bcebos.com/v1/paddle-slim-models/act/ssd_mobilenet_v1_300_120e_voc.tar
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 
 Distillation:
   alpha: 1.0
-  loss: l2
+  loss: soft_label
   node:
   - concat_0.tmp_0
   - concat_2.tmp_0
+  - concat_1.tmp_0
 
 Quantization:
-  activation_quantize_type: 'range_abs_max'
+  use_pact: True
+  weight_quantize_type: 'channel_wise_abs_max'
+  activation_quantize_type: 'moving_average_abs_max'
   quantize_op_types:
   - conv2d
   - depthwise_conv2d
+  onnx_format: True
 
 TrainConfig:
-  train_iter: 80000
-  eval_iter: 1000
-  learning_rate:  
-    type: CosineAnnealingDecay
-    learning_rate: 0.00001
-    T_max: 120000
+  epochs: 5
+  eval_iter: 300
+  learning_rate: 0.001
   optimizer_builder:
     optimizer: 
       type: SGD
     weight_decay: 4.0e-05
-
+  origin_metric: 0.738
diff --git a/example/auto_compression/image_classification/README.md b/example/auto_compression/image_classification/README.md
@@ -44,7 +44,7 @@
 | MobileNetV3_large_x1_0 | Baseline | 75.32 | - | 16.62 | - | [Model](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/MobileNetV3_large_x1_0_infer.tar) |
 | MobileNetV3_large_x1_0 | 量化+蒸馏 | 74.04 | - | 9.85 | [Config](./configs/MobileNetV3_large_x1_0/qat_dis.yaml) | [Model](https://paddle-slim-models.bj.bcebos.com/act/MobileNetV3_large_x1_0_QAT.tar) |
 | MobileNetV3_large_x1_0_ssld | Baseline | 78.96 | - | 16.62 | - | [Model](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/MobileNetV3_large_x1_0_ssld_infer.tar) |
-| MobileNetV3_large_x1_0_ssld | 量化+蒸馏 | 77.17 | - | 9.85 | [Config](./configs/MobileNetV3_large_x1_0/qat_dis.yaml) | [Model](https://paddle-slim-models.bj.bcebos.com/act/MobileNetV3_large_x1_0_ssld_QAT.tar) |
+| MobileNetV3_large_x1_0_ssld | 量化+蒸馏 | 78.17 | - | 9.85 | [Config](./configs/MobileNetV3_large_x1_0/qat_dis.yaml) | [Model](https://paddle-slim-models.bj.bcebos.com/act/MobileNetV3_large_x1_0_ssld_QAT.tar) |
 
 - ARM CPU 测试环境：`SDM865(4xA77+4xA55)`
 - Nvidia GPU 测试环境：