Khadas Docs

Amazing Khadas, always amazes you!

User Tools

Site Tools

Home / products / sbc / edge2 / npu / demos / densenet-ctc
products:sbc:edge2:npu:demos:densenet-ctc

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Next revision 		Previous revision
products:sbc:edge2:npu:demos:densenet-ctc [2023/08/22 06:05]
louis created 		products:sbc:edge2:npu:demos:densenet-ctc [2025/04/09 23:25] (current)
louis
Line 1: Line 1:
-====== Demo3 densenet_ctc ======+~~tag> DenseNet NPU Edge2 RK3588~~
  
-===== Get Source Code =====+====== DenseNet CTC ONNX Keras Edge2 Demo - 3 ======
  
-The codes we use.+{{indexmenu_n>3}} 
 + 
 +===== Introduction ===== 
 + 
 +Densenet_CTC is a text recognition model. It only can recognize single line text. Therefore usually, it needs to be used in conjunction with a text detection model. 
 + 
 +Recognition image and inference results on Edge2. 
 + 
 +{{:products:sbc:edge2:npu:demos:khadasteam.png?400|}} {{:products:sbc:edge2:npu:demos:densenet-ctc-result.png?400|}} 
 + 
 + 
 +===== Train Model ===== 
 + 
 +We will use a DenseNet model based on [[gh>YCG09/chinese_ocr]].
  
 ```shell ```shell
Line 34: Line 47:
 ==== Get convert tool ==== ==== Get convert tool ====
  
-Download Tool from [[https://github.com/rockchip-linux/rknn-toolkit2.git|Rockchip Github]].+Download Tool from [[gh>rockchip-linux/rknn-toolkit2]].
  
 ```shell ```shell
Line 41: Line 54:
 ``` ```
  
-Install dependences and RKNN toolkit2 packages,+Install dependences and RKNN toolkit2 packages.
  
 ```shell ```shell
Line 51: Line 64:
 ``` ```
  
-==== convert ====+==== Convert ====
  
 After training model, run the codes as follows to modify net input and output and convert model to onnx. After training model, run the codes as follows to modify net input and output and convert model to onnx.
  
 <WRAP tip > <WRAP tip >
-Keras model(.h5) can convert rknn model directly. If you want to convert keras model, please use ‘’model.save’’ to save model with weight and network structure.+Keras model(.h5) can convert rknn model directly. If you want to convert keras model, please use ''model.save'' to save model with weight and network structure.
 </WRAP> </WRAP>
  
  
-```shell+```python export.py
 import onnx import onnx
 from keras.models import * from keras.models import *
Line 81: Line 94:
 ``` ```
  
-Enter rknn-toolkit2/examples/onnx/yolov5 and modify test.py as follows.+Enter ''rknn-toolkit2/examples/onnx/yolov5'' and modify ''test.py'' as follows.
  
-```shell+```python test.py
 # Create RKNN object # Create RKNN object
 rknn = RKNN(verbose=True) rknn = RKNN(verbose=True)
Line 94: Line 107:
 # Load ONNX model # Load ONNX model
 print('--> Loading model') print('--> Loading model')
-ret = rknn.load_onnx(model=”./densenet_ctc.onnx”)+ret = rknn.load_onnx(model='./densenet_ctc.onnx')
 if ret != 0: if ret != 0:
     print('Load model failed!')     print('Load model failed!')
Line 102: Line 115:
 # Build model # Build model
 print('--> Building model') print('--> Building model')
-ret = rknn.build(do_quantization=True, dataset=”./dataset.txt”)+ret = rknn.build(do_quantization=True, dataset='./dataset.txt')
 if ret != 0: if ret != 0:
     print('Build model failed!')     print('Build model failed!')
Line 110: Line 123:
 # Export RKNN model # Export RKNN model
 print('--> Export rknn model') print('--> Export rknn model')
-ret = rknn.export_rknn(“./densenet_ctc.rknn”)+ret = rknn.export_rknn('./densenet_ctc.rknn')
 if ret != 0: if ret != 0:
     print('Export rknn model failed!')     print('Export rknn model failed!')
Line 117: Line 130:
 ``` ```
  
-Run test.py to generate rknn model.+Run ''test.py'' to generate rknn model.
  
 ```shell ```shell
Line 127: Line 140:
 ==== Get source code ==== ==== Get source code ====
  
-Clone the source code form our [[https://github.com/khadas/edge2-npu|Github]].+Clone the source code from our [[gh>khadas/edge2-npu]].
  
 ```shell ```shell
Line 144: Line 157:
 === Picture input demo === === Picture input demo ===
  
-Put densenet_ctc.rknn in edge2-npu/C++/densenet_ctc/data/model.+Put ''densenet_ctc.rknn'' in ''edge2-npu/C++/densenet_ctc/data/model''.
  
 ```shell ```shell
-// compile+# Compile 
 +$ cd edge2-npu/C++/densenet_ctc
 $ bash build.sh $ bash build.sh
  
-// run+# Run
 $ cd install/densenet_ctc $ cd install/densenet_ctc
 $ ./densenet_ctc data/model/densenet_ctc.rknn data/img/KhadasTeam.png $ ./densenet_ctc data/model/densenet_ctc.rknn data/img/KhadasTeam.png
Line 156: Line 170:
  
 <WRAP tip > <WRAP tip >
-If your yolov8n model classes is not the same as coco, please change ‘’data/class_str.txt’’ and the ‘’OBJ_CLASS_NUM’’ in ‘’include/postprocess.h’’.+If your **DenseNet CTC** model classes is not the same as mine, please change ''data/class_str.txt'' and the ''OBJ_CLASS_NUM'' in ''include/postprocess.h''.
 </WRAP> </WRAP>
  
Last modified: 2023/08/22 06:05 by louis

Page Tools

Khadas Site

Khadas Community

Khadas GitHub

Khadas Downloads