uci-carl / motionenergy Goto Github PK

The cudaMalloc allocation size is wrong for d_v1GausBuf. It should scale with v1GaussFiltSize, not with nrT.

Add doxygen documentation.

motion_energy.h: No such file or directory.how to solve it?

Add a C++ public interface that honors pImpl and integrates with the core CUDA code.

The cudaMalloc allocation size is wrong for d_v1GausBuf. It should scale with v1GaussFiltSize, not with nrT.

Instead of decreasing the activity of neurons coding for locations close to the border, add an option that paddes the image with zeros before blurring. Look to original S&H Matlab implementation.

There should be a way to record and retrieve execution time.

There should be a test case that benchmarks performance, such as execution time and memory usage.

Add a gtest testing suite.

The directory structure should be reorganized to allow for different front ends (C++ and Python so far).
Change as follows.

.
├── AUTHORS
├── C++.gitignore
├── CHANGELOG
├── cppME
│   ├── a.out
│   ├── doc
│   ├── include
│   │   ├── cuda_version_control.h
│   │   └── motion_energy.h
│   ├── lib
│   ├── libME.a.1.0.0
│   ├── libME.mk
│   ├── Makefile
│   ├── motion_energy.o
│   ├── projects
│   │   ├── include
│   │   ├── main
│   │   ├── Makefile
│   │   ├── results
│   │   └── src
│   │       └── main.cpp
│   └── src
│       └── motion_energy.cu
├── LICENSE
├── pyME
│   ├── doc
│   ├── pyME
│   │   ├── __init__.py
│   │   ├── __main__.py
│   │   └── test
│   │       ├── __init__.py
│   │       └── test_main.py
│   ├── README
│   └── setup.py
├── Python.gitignore
├── README.md
└── RELEASE_NOTES

The new frame should be inserted into d_scalingStimBuf only after the existing frames have been shifted by 1 in time.

Import the existing code from CARLsim 2.1 as release version 0.1

In the hello world example, using a fixed constant for the srand() function doesn't show the same results across multiple runs. Is this expected ?
run 1:
0: mean activity = 0.439031
1: mean activity = 0.501078
2: mean activity = 0.317878
3: mean activity = 0.729056
4: mean activity = 0.502201
5: mean activity = 0.625352
6: mean activity = 0.563613
7: mean activity = 0.688112
8: mean activity = 0.438253
9: mean activity = 0.688300
run 2:
0: mean activity = 0.500670
1: mean activity = 0.563959
2: mean activity = 0.568377
3: mean activity = 0.504105
4: mean activity = 0.440226
5: mean activity = 0.625791
6: mean activity = 0.688198
7: mean activity = 0.562702
8: mean activity = 0.688510
9: mean activity = 0.563258

Thanks

MotionEnergy should be compiled as a library.

Add a Python public interface that honors pImpl and integrates with the core CUDA code (use PyCUDA).