search

bushuhui / pi-cnn

High performance convolutional neural network toolbox for C++
7 stars 4 forks source link

PIL package make error #1

Open amiltonwong opened 8 years ago

amiltonwong commented 8 years ago

Hi, @bushuhui , I had followed the steps you specified, but came across PIL package make error. Build log as follows:

root@milton-Desktop:~/pi-cnn/Thirdparty/PIL# make Compiling librarys of PIL make -j1 -C src make[1]: Entering directory /root/pi-cnn/Thirdparty/PIL/src' make[2]: Entering directory/root/pi-cnn/Thirdparty/PIL/src/base' make -j1 -f ../Makefile.lib make[3]: Entering directory /root/pi-cnn/Thirdparty/PIL/src/base' types/SIM3.h:5:23: fatal error: TooN/TooN.h: No such file or directory compilation terminated. make[3]: *** No rule to make target/root/pi-cnn/Thirdparty/PIL/build/src/base/types/SIM3.d', needed by /root/pi-cnn/Thirdparty/PIL/build/src/base/types/SIM3.o'. Stop. make[3]: Leaving directory/root/pi-cnn/Thirdparty/PIL/src/base' make[2]: * [all] Error 2 make[2]: Leaving directory `/root/pi-cnn/Thirdparty/PIL/src/base' make[1]: * [all] Error 2 make[1]: Leaving directory`/root/pi-cnn/Thirdparty/PIL/src' make: *\ [libs] Error 2

Could you suggest me how to fix it?

(My system is Ubuntu 12.04 (64bit)) Thanks in advance~ Milton

bushuhui commented 8 years ago

Please install TooN before compile the code. TooN is one of code wrote by the author of PTAM. You can find the code through searching PTAM.

amiltonwong commented 8 years ago

Thanks for your quick reply. I installed Toon as guided from : https://github.com/edrosten/TooN Then I can build PIL and pi-cnn with no error. :) Here is the log from ./test_CNN useGPU=1 https://gist.github.com/amiltonwong/506ad78523a91e1b50a1

One quesiton: From the log, I see that the input dimension for CNN model is 224x224x3, but the resolution of original RGB image (./data/84.png and ./data/127.png) is 960x540 . Did the code firstly resize the input image into lower resolution?

bushuhui commented 8 years ago

The advantage of the code is that it divide the input image into several 224x224x3 images, and then compute the feature map separately. Finally, combine them together to form the feature map as the same size of the input image. So you do not need to do the pre-processing.