A2NN-TEST

The project tests the A2NN and Q-A2NN for optical remote sensing classification.

test_code: 
  the test code for comparing the CNN, A2NN and Q-A2NN for optical remote sensing classification.
  
hardware:
  adder_PE_code:
    the code of the adder kernel processing energy.
  para_rom:
    the parameter files for on-chip ROM initialization when deploying QA2NN on FPGA.
  result_count:
    output result of deploying QA2NN quantized in 4bit on FPGA.
    
vedio_demo:
    demonstration of the entire process of test_code and deploying QA2NN quantized in 4bit on FPGA.

---

How to run the ./test_code:

First of all, you need to build a cuda computing environment for the adder kernel. We provide two methods:

· The first method:

We provide configured virtual environments. You only need to download and unzip it, and you can run the test code directly in this environment：

  链接：https://pan.baidu.com/s/1OFKVO-Q5sfJlCNCMygeUDA?pwd=co8a 提取码：co8a 

· The second method:

You can build the environment by youself as follow steps:

(1) Environment Requirements：python 3.8, pytorch 1.8.2, CUDA 11.1

(2) modify PyTorch before launch (for solving compiling issue):

Change lines:57-64 in anaconda3/lib/python3.8/site-packages/torch/include/THC/THCTensor.hpp from:

      #include <THC/generic/THCTensor.hpp>
      #include <THC/THCGenerateAllTypes.h>

      #include <THC/generic/THCTensor.hpp>
      #include <THC/THCGenerateBoolType.h>

      #include <THC/generic/THCTensor.hpp>
      #include <THC/THCGenerateBFloat16Type.h>

to:

      #include <THC/generic/THCTensor.h>
      #include <THC/THCGenerateAllTypes.h>

      #include <THC/generic/THCTensor.h>
      #include <THC/THCGenerateBoolType.h>

      #include <THC/generic/THCTensor.h>
      #include <THC/THCGenerateBFloat16Type.h>

(3) Replace the adder.py with ./adder/adder.py.

Secondly, you can find the CNN, A2NN, and QA2NN versions of the vgg11 pre-trained model from the links we provide, and put these pre-trained models into the test_code folder.

 链接：https://pan.baidu.com/s/1kPuayjwvg---K6XGuJpRjw?pwd=7yqn 提取码：7yqn 

Thirdly, download UC Merced Land Use Dataset and modify the image file path in the /data_process/UC_test_0.2.txt file according to the storage location of the UC dataset.

http://weegee.vision.ucmerced.edu/datasets/landuse.html

After above steps, you can run the test_code:

python test.py

---

The result of deploying QA2NN quantized in 4bit on FPGA:

We provid the output results on the FPGA, you can view the processing results by running the result_count.py:

cd hardware/result_count/
python result.count.py

---

For visual viewing and inspection, we provide test vedio of QA2NN while inferring on GPU and FPGA:

GPU:

链接：https://pan.baidu.com/s/1R4GZbmIDWgJ6SvXbvSJDJA?pwd=ss6j 提取码：ss6j 

FPGA:

链接：https://pan.baidu.com/s/1uZvTiVrFtP84uFwSx8Afrw?pwd=nb4d 提取码：nb4d 

---

Notes: The file size of QA2NN does not reflect the real resource overhead when deploying the inference phase of the QA2NN. We additionally provide the parameter files for on-chip ROM initialization when deploying QA2NN on FPGA.