CFU Playground (2) - CNN - tensorflow lite model

-
 It is not very difficult to change the code from Pytorch to Tensorflow one-to-one, but there are restrictions to use Tensorflow lite micro. Tensorflow lite micro provides various libraries, utilities, and examples in C++ so that it can be run on a microcontroller using ARM core or Tensillica. However, not all neural networks are possible. What is definitely available is a CNN, and a basic fully connected network library is possible. LSTM is on the Kernel, but when converting to Tensorflow lite converter, it was confirmed that errors occurred. Separately, I have seen Github of a person who achieved a degree by implementing LSTM in C. (https://github.com/lephong/lstm-rnn) If you look closely, you can refer to it, but in order to minimize uncertainty, I choose to convert the RNN model into a CNN model.

Source code: 

After changing the times series sequence to a 6x6 array, it look like an image pattern and run it using CNN. Input 0~35, 1~36, 2~37,... , make the input 6x6 and match the y value with the values of 36, 37, 38,... respectively. You may use Conv1D, but Conv2D is advantageous to use Tensorflow lite micro smoothly. The following models can achieve sufficiently similar performance.

Model: "sequential_2" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= conv2d_2 (Conv2D) (None, 6, 6, 4) 8 flatten_2 (Flatten) (None, 144) 0 dense_4 (Dense) (None, 4) 580 dense_5 (Dense) (None, 1) 5 ================================================================= Total params: 593 Trainable params: 593 Non-trainable params: 0 _________________________________________________________________

There is no big problem in making the model, and here you have to first convert it to tensorflow lite without quantization.

# Convert the model to the TensorFlow Lite format without quantization converter = tf.lite.TFLiteConverter.from_saved_model(MODEL_TF) model_no_quant_tflite = converter.convert()

Next, basic optimization is applied to tensorflow lite, and if you want to quantize more, you can create a model by adding a few more options. (Performance is not shown here, and since the board to be applied basically supports floating point, so additional quantize was not applied here. You can try uncomment the some lows in the script)

# Convert the model to the TensorFlow Lite format with quantization converter = tf.lite.TFLiteConverter.from_saved_model(MODEL_TF) converter.optimizations = [tf.lite.Optimize.DEFAULT] #def representative_dataset(): # for value in X_train: # Each scalar value must be inside of a 2D array that is wrapped in a list # yield [np.array(value, dtype = np.float32, ndmin = 4)] # Set the optimization flag. # Enforce integer only quantization # converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8] # converter.inference_input_type = tf.int8 # converter.inference_output_type = tf.int8 # Provide a representative dataset to ensure we quantize correctly. #converter.representative_dataset = representative_dataset model_tflite = converter.convert()

Afterwards, load the saved models to compare the performance, and if you are satisfied with the performance to a certain extent, change it to C++ code with the command below. If you link the generated code with the tensorflow lite micro library, you can run it on the Microcontroller board.

# Convert to a C source file, i.e, a TensorFlow Lite for Microcontrollers model !xxd -i {MODEL_TFLITE} > {MODEL_TFLITE_MICRO} # Update variable names REPLACE_TEXT = MODEL_TFLITE.replace('/', '_').replace('.', '_') !sed -i 's/'{REPLACE_TEXT}'/g_model/g' {MODEL_TFLITE_MICRO}


Comments

Post a Comment

Popular posts from this blog

RNN (5) - FPGA System Design

-
Image
This is the last post on hardware implementation. The RTL codes of the previous post were designed for HW Accelerator IP to perform RNN, and this post describes how to design the whole system for targetting FPGA. The FPGA board used here is an FPGA Board using Xilinx ZYNQ-7000 as open hardware called Pynq-z2. (For Pynq-z2, please refer to the link below.) -  http://www.pynq.io/ -  https://www.tulembedded.com/FPGA/ProductsPYNQ-Z2.html -  https://pynq.readthedocs.io/en/latest/ First, the RNN Accelerator must open a channel to communicate with the ARM-Cortex A9 where SW is running. IP Packaging is performed using Vivado of Xilinx, and I used the AXI interface template provided by Vivado. RTL codes  on the right of the figure below  in the code window  shows how to  combine RNN IP with AXI interface. IP Packaging is possible easily through a few steps in Vivado. (Details will be posted separately.) After IP Packaging, you need to create a bit file to downl...
Read more

RNN (3) - System Simulator

-
Image
The Pytorch reference in the previous post uses the optimized library and floating point number, so the performance can be considered the most ideal. In the system simulator for Verilog design, some restrictions follow.  - Fixed Point Number: Because the range of numbers is limited, performance restrictions follow.  - Suboptimal Matrix Operation: The existing Pytorch model has been optimized for efficient and accurate operation in the Python environment. There is a difference in performance in the System Simulator that is actually implemented.  - Non-linear activation function: Here, tanh (hyperbolic tangent) is used, and linear approximation is performed for convenience of implementation.  Here, a 16-bit fixed point number is used. In fact, all 16 bits are not used and only 7 decimal places are used for easy calculation. Unused redundant bits should be optimized and removed in the future, but the truncation so that prevent overflow during every Addition/Multiplicati...
Read more

RNN (4) - RTL Design

-
Image
I designed the RTL in the same way as the System Simulator designed in the previous post. The major considerations for RTL design are as follows.  (The detailed code description will be posted in a separate post later.  Also, there are many good tutorials on how to use Intel Quartus Modelsim, so you'd better search for it. I will post the explanation of simulating this code in a separate post. ) - I/O method: The most common method is to save the input in the memory, drive the RNN, save the predicted value to output memory, and then send the completion signal. At this time, the memory can be selected in various ways. Various methods such as FIFO, Single Port, Dual Port, etc. can be used. Here, Single Port Memory is selected as the simplest.  - Timing: RTL stands for Register Transfer Level. I don't know who put it on, but it seems to have been put together relatively well. Register is a component whose output is changed at the edge of the clock. Since there is some logic ...
Read more