[cherry-pick]add picodet quant model (#5618)

20b2f101 · Guanghua Yu · GitHub · 6b525e00 · 20b2f101 · 20b2f101
3 changed file
--- a/configs/picodet/README.md
+++ b/configs/picodet/README.md
@@ -219,9 +219,10 @@ paddle2onnx --model_dir output_inference/picodet_s_320_coco_lcnet/ \
 ### 部署
- PaddleInference demo [Python](../../deploy/python) & [C++](../../deploy/cpp)
+- OpenVINO demo [Python](../../deploy/third_engine/demo/openvino/python)
 - [PaddleLite C++ demo](../../deploy/lite)
 - [Android demo(Paddle Lite)](https://github.com/PaddlePaddle/Paddle-Lite-Demo/tree/develop/object_detection/android/app/cxx/picodet_detection_demo)
+- PaddleInference demo [Python](../../deploy/python) & [C++](../../deploy/cpp)
 Android demo可视化：
@@ -236,43 +237,35 @@ Android demo可视化：
 <summary>依赖包:</summary>
 - PaddlePaddle >= 2.2.2
- PaddleSlim >= 2.2.1
+- PaddleSlim >= 2.2.2
 **安装:**
 ```shell
-pip install paddleslim==2.2.1
+pip install paddleslim==2.2.2
 ```
 </details>
-<details>
+<details open>
-<summary>量化训练 (点击展开)</summary>
+<summary>量化训练</summary>
 开始量化训练:
 ```shell
-python tools/train.py -c configs/picodet/picodet_s_320_coco_lcnet.yml \
+python tools/train.py -c configs/picodet/picodet_s_416_coco_lcnet.yml \
-          --slim_config configs/slim/quant/picodet_s_quant.yml --eval
+          --slim_config configs/slim/quant/picodet_s_416_lcnet_quant.yml --eval
 ```
 - 更多细节请参考[slim文档](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/slim)
 </details>
-<details>
+- 量化训练Model ZOO：
-<summary>离线量化 (点击展开)</summary>
-校准及导出量化模型:
-```shell
+| 量化模型     | 输入尺寸 | mAP<sup>val<br>0.5:0.95  | Configs | Weight | Inference Model | Paddle Lite(INT8) |
-python tools/post_quant.py -c configs/picodet/picodet_s_320_coco_lcnet.yml \
+| :-------- | :--------: | :--------------------: | :-------: | :----------------: | :----------------: | :----------------: |
-          --slim_config configs/slim/post_quant/picodet_s_ptq.yml
+| PicoDet-S |  416*416   |  31.5  | [config](./picodet_s_416_coco_lcnet.yml) &#124; [slim config](../slim/quant/picodet_s_416_lcnet_quant.yml) | [model](https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet_quant.pdparams)  | [w/ 后处理](https://paddledet.bj.bcebos.com/deploy/Inference/picodet_s_416_coco_lcnet_quant.tar) &#124; [w/o 后处理](https://paddledet.bj.bcebos.com/deploy/Inference/picodet_s_416_coco_lcnet_quant_non_postprocess.tar) |  [w/ 后处理](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_coco_lcnet_quant.nb) &#124; [w/o 后处理](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_coco_lcnet_quant_non_postprocess.nb) |
-```
- 注意: 离线量化模型精度问题正在解决中.
-</details>
 ## 非结构化剪枝

--- a/configs/picodet/README_en.md
+++ b/configs/picodet/README_en.md
@@ -216,9 +216,10 @@ paddle2onnx --model_dir output_inference/picodet_s_320_coco_lcnet/ \
 ### Deploy
- PaddleInference demo [Python](../../deploy/python) & [C++](../../deploy/cpp)
+- OpenVINO demo [Python](../../deploy/third_engine/demo/openvino/python)
 - [PaddleLite C++ demo](../../deploy/lite)
 - [Android demo(Paddle Lite)](https://github.com/PaddlePaddle/Paddle-Lite-Demo/tree/develop/object_detection/android/app/cxx/picodet_detection_demo)
+- PaddleInference demo [Python](../../deploy/python) & [C++](../../deploy/cpp)
 Android demo visualization:
@@ -233,43 +234,35 @@ Android demo visualization:
 <summary>Requirements:</summary>
 - PaddlePaddle >= 2.2.2
- PaddleSlim >= 2.2.1
+- PaddleSlim >= 2.2.2
 **Install:**
 ```shell
-pip install paddleslim==2.2.1
+pip install paddleslim==2.2.2
 ```
 </details>
-<details>
+<details open>
-<summary>Quant aware (click to expand)</summary>
+<summary>Quant aware</summary>
 Configure the quant config and start training:
 ```shell
-python tools/train.py -c configs/picodet/picodet_s_320_coco_lcnet.yml \
+python tools/train.py -c configs/picodet/picodet_s_416_coco_lcnet.yml \
-          --slim_config configs/slim/quant/picodet_s_quant.yml --eval
+          --slim_config configs/slim/quant/picodet_s_416_lcnet_quant.yml --eval
 ```
 - More detail can refer to [slim document](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/slim)
 </details>
-<details>
+- Quant Aware Model ZOO：
-<summary>Post quant (click to expand)</summary>
-Configure the post quant config and start calibrate model:
-```shell
+| Quant Model     | Input size | mAP<sup>val<br>0.5:0.95  | Configs | Weight | Inference Model | Paddle Lite(INT8) |
-python tools/post_quant.py -c configs/picodet/picodet_s_320_coco_lcnet.yml \
+| :-------- | :--------: | :--------------------: | :-------: | :----------------: | :----------------: | :----------------: |
-          --slim_config configs/slim/post_quant/picodet_s_ptq.yml
+| PicoDet-S |  416*416   |  31.5  | [config](./picodet_s_416_coco_lcnet.yml) &#124; [slim config](../slim/quant/picodet_s_416_lcnet_quant.yml)  | [model](https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet_quant.pdparams)  | [w/ postprocess](https://paddledet.bj.bcebos.com/deploy/Inference/picodet_s_416_coco_lcnet_quant.tar) &#124; [w/o postprocess](https://paddledet.bj.bcebos.com/deploy/Inference/picodet_s_416_coco_lcnet_quant_non_postprocess.tar) |  [w/ postprocess](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_coco_lcnet_quant.nb) &#124; [w/o postprocess](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_coco_lcnet_quant_non_postprocess.nb) |
-```
- Notes: Now the accuracy of post quant is abnormal and this problem is being solved.
-</details>
 ## Unstructured Pruning

--- a/configs/slim/quant/picodet_s_416_lcnet_quant.yml
+++ b/configs/slim/quant/picodet_s_416_lcnet_quant.yml
+pretrain_weights: https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet.pdparams
+slim: QAT
+QAT:
+  quant_config: {
+    'activation_preprocess_type': 'PACT',
+    'weight_quantize_type': 'channel_wise_abs_max', 'activation_quantize_type': 'moving_average_abs_max',
+    'weight_bits': 8, 'activation_bits': 8, 'dtype': 'int8', 'window_size': 10000, 'moving_rate': 0.9,
+    'quantizable_layer_type': ['Conv2D', 'Linear']}
+  print_model: False
+TrainReader:
+  batch_size: 48
+LearningRate:
+  base_lr: 0.024
+  schedulers:
+  - !CosineDecay
+    max_epochs: 300
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 300