I try to work with mask RCNN and I have some problems with the first example that exist: kangaroo. By train I get an error that the data are boolean:
ValueError: Input image dtype is bool. Interpolation is not defined with bool data type. Please set order to 0 or explicitely cast input image to another data type.
Second error or warning is:
W tensorflow/compiler/mlir/tools/kernel_gen/tf_gpu_runtime_wrappers.cc:30] ‘cuModuleUnload(it.second)’ failed with ‘CUDA_ERROR_NOT_INITIALIZED’
The system that I use have following attributes:
ubuntu: 22.04.2 LTS
Graphic Card: NVIDIA GeForce RTX 3090
nvidia driver: 530.20.02
CUDA 12.1
CuDNN 8.8.1
Tensorflow 2.8.0
As start I intall MASK_RCNN_TF2 from git What should I do to make it working.
I install all the componets at lest 2 times and I try also Tensorflow 2.11, 2.5.0, 2.8.0. Same results
@Paul-Tiberiu_Miclea,
Welcome to the Tensorflow Forum!
ValueError: Input image dtype is bool. Interpolation is not defined with bool data type. Please set order to 0 or explicitely cast input image to another data type.
You can try to cast the input image to another data type before passing it to the model.
tensorflow/compiler/mlir/tools/kernel_gen/tf_gpu_runtime_wrappers.cc:30] ‘cuModuleUnload(it.second)’ failed with ‘CUDA_ERROR_NOT_INITIALIZED’
Could you please try as per tested build configurations : CUDA 11.2 and cuDNN : 8.1 ?
We recommend to use official supported tfm.vision.MaskRCNNTask -Provides artifacts for training/evalution procedures, including loading/iterating over Datasets, initializing the model, calculating the loss, post-processing and customized metrics with reduction.
Thank you!
Thank you very much. As I look on NVidia the Cuda 11.x is not available for ubuntu 22.04, ant this was the reason way I have tu use CUDA 12. Maybe I’m wong. The Images that I use are 8bit gray scale images (normally from a SEM).
Because I am new in Tensorflow, I don’t know that exist( sorry!). I will try. Exist a documentation or exaples how to start with? Are the Datasets comparable with the other methods: labes in json format?
Thank you
@Paul-Tiberiu_Miclea,
Please take a look at the Instance Segmentation with Model Garden | TensorFlow Core tutorial to fine-tunes a Mask R-CNN with Mobilenet V2 as the backbone model.
Also, please take a look at tfm.core.exp_factory.get_exp_config | TensorFlow v2.11.0 for all the registered experiments.
Thank you!