test pretrain model with customized data

[Anguliachao]

Hi @Engineering-Course ,
thank you for the great project, I''ve successfully compiled ur caffe version and try to execute run_human.sh, but it prompts the error in image transformer processing.
data_transformer.cpp:474] Check failed: img_channels == data_channels (1 vs. 3)
However I follow your instruction to make a val.txt to construct my testing image list.
the sample in my txt file such as
/images/23454.jpg
/images/3534.jpg
all of them are 3 channel RGB images, but still I got this error , may I ask about the reason? seems like img_channels should also be 3?
Thank you if you can check this for me.

[Anguliachao]

The full message as follows:
./run_human.sh Testing net human/attention Running /XXX/caffe_ssl/.build_release/tools/caffe.bin test --model=human/config/attention/test_val.prototxt --weights=human/model/attention/test.caffemodel --gpu=0 --iterations=143 F1205 09:28:22.869490 14228 data_transformer.cpp:474] Check failed: img_channels == data_channels (1 vs. 3) *** Check failure stack trace: *** @ 0x7f91ac8895cd google::LogMessage::Fail() @ 0x7f91ac88b433 google::LogMessage::SendToLog() @ 0x7f91ac88915b google::LogMessage::Flush() @ 0x7f91ac88be1e google::LogMessageFatal::~LogMessageFatal() @ 0x7f91acf124fa caffe::DataTransformer<>::TransformImgAndSeg() @ 0x7f91ad0403d0 caffe::ImageSegDataLayer<>::load_batch() @ 0x7f91acff312f caffe::BasePrefetchingDataLayer<>::InternalThreadEntry() @ 0x7f91ad0f2e85 caffe::InternalThread::entry() @ 0x7f91a25cc5d5 (unknown) @ 0x7f91a1e7a6ba start_thread @ 0x7f91ab42441d clone @ (nil) (unknown) Aborted (core dumped)
As for me , during test, we should only read RGB images correct? No need to read Seg Img, How exactly you made the val.txt and val_id.txt

[Engineering-Course]

Sorry, but the current code still needs Seg Img during testing.
In another word, the txt file should be like /image/23454.jpg /segmentation/23454.jpg.
You may generate an empty Seg Img for your data to suit the input.
Or you can revise the ImageSegData layer.

[Anguliachao]

@Engineering-Course Ok sure , will revice the data layer ,thank you.