RKNN Toolkit: update version to 1.7.5.

Signed-off-by: raul.rao <raul.rao@rock-chips.com>

README.md

@@ -11,32 +11,37 @@ Note:
 # Download
 From version 1.3.0, some wheel packages are larger than 100MB, can not be uploaded directly. So, you need to go to the releases page to download.
 You can download from releases page: https://github.com/rockchip-linux/rknn-toolkit/releases
-- All wheel packages are in compressed file: [rknn-toolkit-v1.7.3-packages.tar.gz](https://github.com/rockchip-linux/rknn-toolkit/releases/download/v1.7.3/rknn-toolkit-v1.7.3-packages.tar.gz "rknn-toolkit-v1.7.3-packages.tar.gz") or [rknn-toolkit-v1.7.3-packages.zip](https://github.com/rockchip-linux/rknn-toolkit/releases/download/v1.7.3/rknn-toolkit-v1.7.3-packages.zip "rknn-toolkit-v1.7.3-packages.zip ")
-- All examples, docs and platform-tools are in compressed file: [Source code(zip)](https://github.com/rockchip-linux/rknn-toolkit/archive/v1.7.3.zip "Source code(zip)") or [Source code(tar.gz)](https://github.com/rockchip-linux/rknn-toolkit/archive/v1.7.3.tar.gz "Source code(tar.gz)")
+- All wheel packages are in compressed file: [rknn-toolkit-v1.7.5-packages.tar.gz](https://github.com/rockchip-linux/rknn-toolkit/releases/download/v1.7.5/rknn-toolkit-v1.7.5-packages.tar.gz "rknn-toolkit-v1.7.5-packages.tar.gz") or [rknn-toolkit-v1.7.5-packages.zip](https://github.com/rockchip-linux/rknn-toolkit/releases/download/v1.7.5/rknn-toolkit-v1.7.5-packages.zip "rknn-toolkit-v1.7.5-packages.zip ")
+- All examples, docs and platform-tools are in compressed file: [Source code(zip)](https://github.com/rockchip-linux/rknn-toolkit/archive/v1.7.5.zip "Source code(zip)") or [Source code(tar.gz)](https://github.com/rockchip-linux/rknn-toolkit/archive/v1.7.5.tar.gz "Source code(tar.gz)")
 - You can also download all packages, docker image, examples, docs and platform-tools from baidu cloud: [rknn-toolkit](https://eyun.baidu.com/s/3bqgIr0N "RKNN-Toolkit"), fetch code: rknn
 - You can download RKNN Toolkit Lite packages and examples from [rknn-toolkit-lite](rknn-toolkit-lite)
 # Checksums
 ## MD5
 ```
-7d672e911c26af34112bbfcfed86aef7  rknn_toolkit-1.7.3-cp35-cp35m-linux_aarch64.whl
-e751a1c2782879dedbe3904448e07105  rknn_toolkit-1.7.3-cp36-cp36m-linux_x86_64.whl
-20db420731b4976b15b0fbddef9984e3  rknn_toolkit-1.7.3-cp36-cp36m-macosx_10_15_x86_64.whl
-1ebba11b78766cd320a470bb397578cb  rknn_toolkit-1.7.3-cp36-cp36m-win_amd64.whl
-0cdd9c288a748bcef84848fd5dc12d80  rknn_toolkit-1.7.3-cp37-cp37m-linux_aarch64.whl
-f06e1f88864c5a4759e4258cb536c38d  rknn_toolkit-1.7.3-cp37-cp37m-macosx_10_15_x86_64.whl
-8e5c63a241b809b78ca65d226344d0eb  rknn_toolkit-1.7.3-cp38-cp38-linux_x86_64.whl
+055ea6bc3b82cd437ebe645076c66960  rknn_toolkit-1.7.5-cp35-cp35m-linux_aarch64.whl
+dec84a1226a22914e9912f1ce61cad64  rknn_toolkit-1.7.5-cp36-cp36m-linux_x86_64.whl
+3a0f2d0cadb58e60f26fe039394205ab  rknn_toolkit-1.7.5-cp36-cp36m-macosx_10_15_x86_64.whl
+af96a8ab4cffa1037d41deb155e3d92e  rknn_toolkit-1.7.5-cp36-cp36m-win_amd64.whl
+4511b18f5a3127fb7375c2dddf597641  rknn_toolkit-1.7.5-cp37-cp37m-linux_aarch64.whl
+8873a605594264125aa33bdaf2fc08b3  rknn_toolkit-1.7.5-cp37-cp37m-macosx_10_15_x86_64.whl
+da4cd1a1968f9e9aa5414a67598f5374  rknn_toolkit-1.7.5-cp38-cp38-linux_x86_64.whl
 
-a24f157407e16bc255fc387404eb2030  rknn-toolkit-v1.7.3-packages.tar.gz
-8dbeeecb06b9201b9f464909ad8d9544  rknn-toolkit-v1.7.3-packages.zip
+0818a331d3bba755036cfee296bdad31  rknn-toolkit-v1.7.5-packages.tar.gz
+f5de735b9b733d74f97db5cce46c2c70  rknn-toolkit-v1.7.5-packages.zip
 
-4b50d8966908e80567843c3623e33d46  rknn_toolkit_lite-1.7.3-cp36-cp36m-linux_aarch64.whl
-dc669361506646104e823d569055b849  rknn_toolkit_lite-1.7.3-cp36-cp36m-linux_x86_64.whl
-15a26aad303769f8f2f38f5888529a57  rknn_toolkit_lite-1.7.3-cp36-cp36m-macosx_10_15_x86_64.whl
-cd490d13dc1e40e24ab76100a9b82ced  rknn_toolkit_lite-1.7.3-cp36-cp36m-win_amd64.whl
-17f10789ec42fdc687c3c02ef6dc1141  rknn_toolkit_lite-1.7.3-cp37-cp37m-linux_aarch64.whl
-030a09ed522620aa6dfb4ccac578518b  rknn_toolkit_lite-1.7.3-cp37-cp37m-linux_armv7l.whl
-079aaf9c2c39b2c2a6858688ed733973  rknn_toolkit_lite-1.7.3-cp37-cp37m-macosx_10_15_x86_64.whl
-b1b32a4e8803dd9303ab1a72ae8deccf  rknn_toolkit_lite-1.7.3-cp38-cp38-linux_x86_64.whl
+7adbd9698c7b528413d0ba73b591c83f  rknn_toolkit_lite-1.7.5-cp310-cp310-linux_aarch64.whl
+3457be77486bcd70c66213f46ce223e3  rknn_toolkit_lite-1.7.5-cp35-cp35m-linux_aarch64.whl
+5d1ed1ac6dff669be03578ce39787eea  rknn_toolkit_lite-1.7.5-cp36-cp36m-linux_aarch64.whl
+2f15ccb2c4140a436d5dfbc8e9544630  rknn_toolkit_lite-1.7.5-cp36-cp36m-linux_armv7l.whl
+210fe992928bd57ff638d346e394a5f2  rknn_toolkit_lite-1.7.5-cp36-cp36m-linux_x86_64.whl
+16b13b9b710b90f2de10c820180f1c51  rknn_toolkit_lite-1.7.5-cp36-cp36m-macosx_10_15_x86_64.whl
+27012b2aa02b78a3720bf9d34e5a42cf  rknn_toolkit_lite-1.7.5-cp36-cp36m-win_amd64.whl
+90d1f4c19552837a60e7d713c8b86e01  rknn_toolkit_lite-1.7.5-cp37-cp37m-linux_aarch64.whl
+0dbfe4e8fc4a50c95d5f8114e379bce5  rknn_toolkit_lite-1.7.5-cp37-cp37m-linux_armv7l.whl
+82e046302a2527bec6ebbd957446e8e6  rknn_toolkit_lite-1.7.5-cp37-cp37m-macosx_10_15_x86_64.whl
+efe50cbb488c49f4953ff156caef6d07  rknn_toolkit_lite-1.7.5-cp38-cp38-linux_aarch64.whl
+59f3f1df13bad289daadab276684c8df  rknn_toolkit_lite-1.7.5-cp38-cp38-linux_x86_64.whl
+63868f54eae2c98da69679abf4710881  rknn_toolkit_lite-1.7.5-cp39-cp39-linux_aarch64.whl
 ```
 # Feedback and Community Support
 - QQ Group Chat: 1025468710 (full, please join group 2)

doc/changelog.txt

@@ -1,3 +1,23 @@
+2023-07-31
+版本：v1.7.5
+1. 新增功能：
+   增加模型稀疏化功能；
+   增加量化参数导出功能；
+   量化模块增加离群点检测功能；
+   性能分析模块增加各类型算子耗时统计功能。
+2. 功能优化：
+   完善对 PyTorch QAT 模型的支持；
+   重写 ONNX 模型优化器；
+   完善可视化功能；
+   优化 MMSE 功能；
+   完善图优化模块；
+   TFLite 转换器增加对不支持算子的检测。
+3. 完善算子支持, 新增erf支持。
+4. 移除对离线预编译 / 模拟器的支持。
+5. 升级部分依赖模块。
+6. 修复已知bug。
+
+
 2022-08-20
 版本：v1.7.3
 1. 功能优化：

examples/caffe/mobilenet_v2/README.md

@@ -0,0 +1,75 @@
+# Example for Caffe mobilenet_v2
+
+
+## Model Source
+The model used in this example come from the following open source projects:
+https://github.com/shicai/MobileNet-Caffe
+
+
+## Usage for the script
+
+*Usage:*
+```
+python test.py [target] [device_id]
+```
+*Parameter Description:*
+- target: target platform. Optional parameter, the default value is `rv1126`, you can fill in `rk1806`, `rk1808`, `rk3399pro`, `rv1109`, `rv1126`.
+- device_id: Device ID, when multiple devices are connected, this parameter is used to distinguish them. Optional parameter, default value is None.
+
+If the target device is `RV1109` or `RV1126`, you can directly execute the following command to run the example:
+```
+python test.py
+```
+If the target device is RK1806, RK1808 or RK3399Pro, you can execute the following command to run the example:
+```
+python test.py rk1808
+```
+If you connect multiple devices, you need to specify the device ID, please refer to the following command to run the example:
+```
+python test.py rv1126 c3d9b8674f4b94f6
+```
+
+
+## Expected results
+
+This example will print the TOP5 labels and corresponding scores of the test image classification results. For example, the inference results of this example are as follows:
+```
+[1]: 0.73388671875
+[115 996]: 0.033447265625
+[115 996]: 0.033447265625
+[927]: 0.026214599609375
+[794]: 0.024169921875
+```
+
+1. The label index with the highest score is 1, the corresponding label is `goldfish`.
+2. The download link for labels file: https://github.com/HoldenCaulfieldRye/caffe/blob/master/data/ilsvrc12/synset_words.txt.
+3. Different platforms, different versions of tools and drivers may have slightly different results.
+
+
+## Notes
+
+- The prototxt in the open source model uses the old version of Caffe, which has been modified in the actual example. The modified content is as follows
+```
+# Comment or remove the following content:
+input: "data"
+input_dim: 1
+input_dim: 3
+input_dim: 224
+input_dim: 224
+
+# Add the following content
+layer {
+  name: "data"
+  type: "Input"
+  top: "data"
+  input_param {
+    shape {
+      dim: 1
+      dim: 3
+      dim: 224
+      dim: 224
+    }
+  }
+}
+```
+

examples/caffe/mobilenet_v2/test.py

@@ -1,3 +1,5 @@
+import platform
+import sys
 import numpy as np
 import cv2
 from rknn.api import RKNN
@@ -22,16 +24,39 @@ def show_outputs(outputs):
 
 
 if __name__ == '__main__':
+    # Default target and device_id
+    target = 'rv1126'
+    device_id = None
+
+    # Parameters check
+    if len(sys.argv) == 1:
+        print("Using default target rv1126")
+    elif len(sys.argv) == 2:
+        target = sys.argv[1]
+        print('Set target: {}'.format(target))
+    elif len(sys.argv) == 3:
+        target = sys.argv[1]
+        device_id = sys.argv[2]
+        print('Set target: {}, device_id: {}'.format(target, device_id))
+    elif len(sys.argv) > 3:
+        print('Too much arguments')
+        print('Usage: python {} [target] [device_id]'.format(sys.argv[0]))
+        print('Such as: python {} rv1126 c3d9b8674f4b94f6'.format(
+            sys.argv[0]))
+        exit(-1)
 
     # Create RKNN object
     rknn = RKNN()
 
     # Set model config
     print('--> Config model')
-    rknn.config(mean_values=[[103.94, 116.78, 123.68]], std_values=[[58.82, 58.82, 58.82]], reorder_channel='2 1 0')
+    rknn.config(mean_values=[[103.94, 116.78, 123.68]],
+                std_values=[[58.82, 58.82, 58.82]],
+                reorder_channel='2 1 0',
+                target_platform=[target])
     print('done')
 
-    # Load caffe model
+    # Load model (from https://github.com/shicai/MobileNet-Caffe)
     print('--> Loading model')
     ret = rknn.load_caffe(model='./mobilenet_v2.prototxt',
                           proto='caffe',
@@ -62,7 +87,11 @@ def show_outputs(outputs):
     img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
 
     print('--> Init runtime environment')
-    ret = rknn.init_runtime()
+    if target.lower() == 'rk3399pro' and platform.machine() == 'aarch64':
+        print('Run demo on RK3399Pro, using default NPU.')
+        target = None
+        device_id = None
+    ret = rknn.init_runtime(target=target, device_id=device_id)
     if ret != 0:
         print('Init runtime environment failed')
         exit(ret)

examples/caffe/vgg-ssd/README.md

@@ -0,0 +1,70 @@
+# Example for Caffe VGG-SSD
+
+
+## Model Source
+The model used in this example come from the following open source projects:
+https://github.com/weiliu89/caffe/tree/ssd
+The download link provided by this project has expired, please download the model weight from the following network disk link(fetch code is `rknn`):
+https://eyun.baidu.com/s/3jJhPRzo
+
+
+## Usage for the script
+
+*Usage:*
+```
+python test.py [target] [device_id]
+```
+*Parameter Description:*
+- target: target platform. Optional parameter, the default value is `rv1126`, you can fill in `rk1806`, `rk1808`, `rk3399pro`, `rv1109`, `rv1126`.
+- device_id: Device ID, when multiple devices are connected, this parameter is used to distinguish them. Optional parameter, default value is None.
+
+If the target device is `RV1109` or `RV1126`, you can directly execute the following command to run the example:
+```
+python test.py
+```
+If the target device is RK1806, RK1808 or RK3399Pro, you can execute the following command to run the example:
+```
+python test.py rk1808
+```
+If you connect multiple devices, you need to specify the device ID, please refer to the following command to run the example:
+```
+python test.py rv1126 c3d9b8674f4b94f6
+```
+
+
+## Expected results
+
+The test result should be similar to picutre `ref_detect_result.jpg`.
+
+*Note: Different platforms, different versions of tools and drivers may have slightly different results.*
+
+
+## Notes
+
+- The prototxt in the open source model uses the old version of Caffe, which has been modified in the actual example. The modified content is as follows
+```
+# Comment or remove the following content:
+input: "data"
+input_shape {
+  dim: 1
+  dim: 3
+  dim: 300
+  dim: 300
+}
+
+# Add the following content:
+layer {
+  name: "input"
+  type: "Input"
+  top: "data"
+  input_param {
+    shape {
+      dim: 1
+      dim: 3
+      dim: 300
+      dim: 300
+    }
+  }
+}
+```
+- The DetectionOutput layer is also been removed.