I want to train single channel images with cascade RCNN, which need to be modified

[penghaoxiao]

I1026 15:39:55.119289 15373 detection_data_layer.cpp:58] Window data layer:
batch size: 2
cache_images: 0
root_folder: /VOCdevkit/
I1026 15:39:55.119340 15373 detection_data_layer.cpp:142] num: 0 /VOCdevkit/VOC2020/JPEGImages/3_3&18.tiff 1 1000 1000 windows to process
I1026 15:39:55.119726 15373 detection_data_layer.cpp:150] Number of images: 53
I1026 15:39:55.119731 15373 detection_data_layer.cpp:155] class 0 has 0 samples
I1026 15:39:55.119736 15373 detection_data_layer.cpp:155] class 1 has 803 samples
I1026 15:39:55.119740 15373 detection_data_layer.cpp:163] Random aspect shuffling data
I1026 15:39:55.120517 15373 detection_data_layer.cpp:233] output data size: 2,1,1024,1024
I1026 15:39:55.120568 15373 detection_data_layer.cpp:254] output label size 0 : 2,6,64,64
I1026 15:39:55.120611 15373 detection_data_layer.cpp:254] output label size 1 : 2,6,64,64
I1026 15:39:55.120657 15373 detection_data_layer.cpp:254] output label size 2 : 2,6,64,64
I1026 15:39:55.120709 15373 detection_data_layer.cpp:254] output label size 3 : 2,6,64,64
I1026 15:39:55.120752 15373 detection_data_layer.cpp:254] output label size 4 : 2,6,64,64
I1026 15:39:55.120791 15373 detection_data_layer.cpp:254] output label size 5 : 2,6,64,64
I1026 15:39:55.120843 15373 detection_data_layer.cpp:254] output label size 6 : 2,6,64,64
I1026 15:39:55.120882 15373 detection_data_layer.cpp:254] output label size 7 : 2,6,64,64
I1026 15:39:55.120923 15373 detection_data_layer.cpp:254] output label size 8 : 2,6,64,64
I1026 15:39:55.120961 15373 detection_data_layer.cpp:270] output gt boxes size: 1,8,1,1
F1026 15:39:55.120966 15373 detection_data_layer.cpp:281] Check failed: mean_values_.size() == 1 || mean_values_.size() == channels Speci
*** Check failure stack trace: ***
@ 0x7fbba1f655cd google::LogMessage::Fail()
@ 0x7fbba1f67433 google::LogMessage::SendToLog()
@ 0x7fbba1f6515b google::LogMessage::Flush()
@ 0x7fbba1f67e1e google::LogMessageFatal::~LogMessageFatal()
@ 0x7fbba26d1478 caffe::DetectionDataLayer<>::DataLayerSetUp()
@ 0x7fbba2617775 caffe::BasePrefetchingDataLayer<>::LayerSetUp()
@ 0x7fbba2564b5c caffe::Net<>::Init()
@ 0x7fbba25673ae caffe::Net<>::Net()
@ 0x7fbba2571b65 caffe::Solver<>::InitTrainNet()
@ 0x7fbba2572f95 caffe::Solver<>::Init()
@ 0x7fbba25732af caffe::Solver<>::Solver()
@ 0x7fbba275cde1 caffe::Creator_SGDSolver<>()
@ 0x416c5a caffe::SolverRegistry<>::CreateSolver()
@ 0x40e25d train()
@ 0x4097d3 main
@ 0x7fbba0e5f840 __libc_start_main
@ 0x40a1d9 _start
@ (nil) (unknown)

[zhaoweicai]

Could you just pad your single channel image to rgb image?