The cocoapi directory contains data from another github here which contains a necessary setup step for successfully running the Mask-RCNN model.
The MaskRCNN directory contains data from a version of Mask-RCNN that has been modified to work with TensorFlow 2 here.
Additional light modifications have been made to get it to work on my system.
- Install Visual Studio Community (2017) and install the C++ development tools (include 2015 Build tools for cocoapi later)
- Install Cuda Version 10.0, use the express installation option.
- Download CuDNN version 7.4 and put the cudnn files in the lib, bin, and include folders into the lib, bin, and include folders of the Cuda installation (C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0)
- Put the bin, libnvvp, and \extras\CUPTI\lib64 folders on the path. Ensure there is a CUDA_PATH and a CUDA_PATH_Vxx_x in your system variables.
This was done on a RTX 3060
- Install Visual Studio Community (2019) and install the C++ development tools (include 2015 Build tools for cocoapi later)
- Install Cuda Version 11.2, use the express installation option.
- Download CuDNN version 8.1 and put the cudnn files in the lib, bin, and include folders into the lib, bin, and include folders of the Cuda installation (C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1)
- Put the bin, libnvvp, and \extras\CUPTI\lib64 folders on the path. Ensure there is a CUDA_PATH and a CUDA_PATH_Vxx_x in your system variables
- Create a conda environment using python 3.8
- Install the packages outlined in the requirements.txt file (provided the previous setup steps are the same).
- Go into the cocoapi folder and continue until you are in the PythonAPI directory
- Type “python setup.py build_ext install”
- Go back into the MRCNN root directory and use the same command.
To run, enter the Mask_RCNN/samples directoru and run the demo.py file. This is a simple demo that will output a result to the Results folder (its supposed to also have a popup but at this version that has not worked on my system)