cannot import cv2
conda install -c https://conda.anaconda.org/menpo opencv3
cAttributeError: 'NoneType' object has no attribute 'astype'
change the image name from .png to .jpg
[[Node: save/Assign_37 = Assign[T=DT_FLOAT, _class=["loc:@model/encoder/Conv/weights"], use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/cpu:0"](model/encoder/Conv/weights, save/RestoreV2_37)]]
add --do_stereo
NotFoundError (see above for traceback): /media/anotherday/fyl/masterthesis/KITTI/data_scene_flowtraining/image_2/000000_10.jpg [[Node: cond/ReadFile = ReadFile_device="/job:localhost/replica:0/task:0/cpu:0"]]
chang the test image to jpg, because the filenames is training/image_2/000004_10.jpg training/image_3/000004_10.jpg
import caffe Traceback (most recent call last): File "", line 1, in File "/home/anotherday/anaconda3/envs/caffe27/lib/python2.7/site-packages/caffe/init.py", line 1, in from .pycaffe import Net, SGDSolver, NesterovSolver, AdaGradSolver, RMSPropSolver, AdaDeltaSolver, AdamSolver, NCCL, Timer File "/home/anotherday/anaconda3/envs/caffe27/lib/python2.7/site-packages/caffe/pycaffe.py", line 13, in from ._caffe import Net, SGDSolver, NesterovSolver, AdaGradSolver,
ImportError: /home/anotherday/anaconda3/envs/caffe27/lib/python2.7/site-packages/caffe/../../../libcaffe.so.1.0.0: undefined symbol: ZN7leveldb2DB4OpenERKNS_7OptionsERKSsPPS0
SOLUTION: conda install -c defaults protobuf libprotobuf
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
SOLUTION: change if image == NONE: to if image is NODE:
IOError: cannot identify image file 'test.ppm'
change "import Image" to "from PIL import Image"
=======
in windows File "C:\Users\fyl\Anaconda3\envs\tf\lib\site-packages\tensorflow\python\training\saver.py", line 1717, in restore + compat.as_text(save_path)) ValueError: The passed save_path is not a valid checkpoint:
SOLUTION: restore_path = args.checkpoint_path #.split(".")[0]
TypeError: init() got an unexpected keyword argument 'constraint'
Solution: change the tensorflow version from 1.2 to the newerst version just because the name is diff in different version of tf
not_loaded.remove(v.op.name)
ValueError: list.remove(x): x not in list
??? not solved: add continue or break
cuda out of memeory
SOLUTION: config = tf.ConfigProto(allow_soft_placement=True) ## let use 70% memeory of gpu gpu_options=tf.GPUOptionis()
seg_map[masks[n_]> 0,:] = colorize(n_, color=FL.color)
IndexError: boolean index did not match indexed array along dimension 0; dimension is 480 but corresponding boolean dimension is 640
add if num is 1 : seg_map[masks > 0,:] = colorize(num, color=FL.color) elif num > 1: in the maskrcnn inference.py line 90
source env.sh or change name from segim to segimg
error:ldd icp_op.so i see the error like: undefined symbol: cublasDsyr2k_v2, version libcublas.so.9.0 (/usr/local/lib/libtensorflow_framework.so) solve: 1)export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}} 2)export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
error: undefined symbol: _ZN10tensorflow12OpDefBuilder3DocESs (./icp_op.so) solve: delete GLIBCXX_USE_CXX11_ABI=0 in cmakelist file
notice: env
(gcc version 5.4 tensorflow address : in python2.7/dis..package/tensorflow/ but not python3.6)
error: ImportError: dynamic module does not define init function (initicp_op) solve: rename icp_op.so to Icp_op.so
(1) download file from tensorflow/models#5168
(2) sudo apt install libpcl-dev (1.7.2 is fine)
(3) sudo apt install libeigen3-dev (eigen must > 3.3/ my version is 3.3.5)
(4) export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}} export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
(5) delete GLIBCXX_USE_CXX11_ABI=0 in cmakelist file modify the CMakeLists.txt like change the tensorflow include the path and so on
(6) ldd -r icp_op.so find if something else is wrong
(7) copy file libtensorflow_framework.so to /usr/local/lib/
(8) cd models/research/vid2depth/ops mkdir build cd build
(9) cmake -DCMAKE_BUILD_TYPE=Release ..
(10) mv libicp_op.so ../Icp_op.so
bazel build -c opt ops/...
########## vid2depth add icp PCL ################################################################# (1)ERROR: /home/anotherday/vid2depth_git/new/vid2depth/WORKSPACE:29:1: name 'http_archive' is not defined solve: add load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive") IN WORKSPACE
(2)ERROR: name 'new_http_archive' is not defined Analysis of target '//ops:icp_op_kernel' failed; build aborted: no such package '@com_github_pointcloudlibrary_pcl//': The native new_http_archive rule is deprecated. load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive") for a drop-in replacement.
solve: CHANGE bazel version to 0.22 --> 0.18 --> 0.11
(3)ERROR: /home/.cache/bazel/_bazel_yz/0a99992d99badecd9853941fabef3318/external/jpeg/BUILD:126:12: Illegal ambiguous match on configurable attribute "deps" in @jpeg//:jpeg: @jpeg//:k8 @jpeg//:armeabi-v7a
ERROR: cannot find mex.h solve: install linux version's matlab not indeed delete matlab file in flann file ( vid2depth/bazel-vid2depth/external/flann/.. matlab/) delete mpi file in flann ( vid2depth/bazel-vid2depth/external/flann/src/cpp/flann/mpi)
(4) error:boost/mpi/config.hpp:20:17: fatal error: mpi.h: No such file or directory
sudo apt install libopenmpi-dev pip install mpi4py
change to #include "/usr/include/mpi/openmpi/ompi/mpi/cxx/exception.h" change to #include <mpi/mpi.h> change to #include "/usr/include/mpi/mpi.h" 解决：将 mpi.h 改为绝对路径 #include "/opt/intel/compilers_and_libraries_2017.0.098/linux/mpi/intel64/include/mpi.h" copy usr/include/boost and usr/include/mpi
(5)ERROR: /home/anotherday/vid2depth_git/new/vid2depth/ops/BUILD:10:1: C++ compilation of rule '//ops:icp_op_kernel' failed (Exit 1) ops/icp_op_kernel.cc: In member function 'void IcpOp::Compute(tensorflow::OpKernelContext*)': ops/icp_op_kernel.cc:147:27: error: 'int64' has not been declared this](int64 start_row, int64 limit_row) {
(6) ERROR: fatal error: mex.h: No such file or directory
export PATH=$PATH:/usr/local/MATLAB/R2019a/bin sudo rm mex in usr/bin/
delete matlab file in flann file delete mpi file in flann ( vid2depth/bazel-vid2depth/external/flann/src/cpp/flann/mpi)
(7) error: error: bazel-out/k8-opt/bin/external/net_zlib_zlib/_objs/zlib/external/net_zlib_zlib/adler32.pic.o: multiple definition of 'adler32_combine64' /usr/bin/ld.gold: bazel-out/k8-opt/bin/external/zlib_archive/_objs/zlib/external/zlib_archive/adler32.pic.o: previous definition here
delete one of them "delete adler32_combine64.pic.o" not sucessufully solution: in /home/anotherday/vid2depth_git/new/vid2depth/bazel-vid2depth/external/net_zlib_zlib/BUILD file, use # to comment the .c file name, for example: # adler32.c (8) ERROR: could not load icp.o solution: copy the icp_op.o to ops file ERROR: undefined symbol: _ZTIN10tensorflow8OpKernelE SOLUTION: 这是 因为在用g++编译代码的时候没用使用tensorflow_framework.so动态库造成的，在gcc+ 的参数中添加-L /usr/local/lib/python3.5(2.7)/dist-packages/tensorflow export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}} export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}} delete GLIBCXX_USE_CXX11_ABI=0 in cmakelist file
use ldd -r icp_op.so to find the reason
failed call to cuInit: CUDA_ERROR_UNKNOWN tensorflow gpu can not work Solution:sudo apt install nvidia-modprobe
return _setlocale(category, locale)locale.Error: unsupported locale setting sudo dpkg-reconfigure locales
GPU OUT OF MEMORY QUEUE_SIZE = 2000 to QUEUE_SIZE = 20 also change the tensorflow gpu session to allow growth
ERROR: SyntaxError: Missing parentheses in call to 'print' SOLUTION: change python 3.6 to python 2.6
(1) git clone https://github.com/occipital/OpenNI2.git
(2) make
(3) go to /Packaging
(4) conda activate caffe_python2.7
(5) ReleaseVersion.py x64
(6) go to /Final
(7) extract OpenNI-Linux-x64-2.2.tar.bz2
(8) go into OpenNI-Linux-x64-2.2
(9) sudo ./install.sh
(10) cat OpenNIDevEnvironment >> ~/.bashrc source ~/.bashrc (11) compile the openni file
ERROR: usb interface is not supported In Xtion
solution: udo nano /etc/openni/GlobalDefaults.ini comment it and it should work ;UsbInterface=2 reboot
error:hdf5.h no such file or directory
solution: export CPATH="/usr/include/hdf5/serial/" in make file USER_LIBS= -L/usr/include/hdf5/serial/ -lhdf5
(1) cp data /input cp sourcecode /input
(2) install pcl
conda install -c conda-forge pcl
(3) change the CMakeLists.txt the tensorflow path (conda info -e)
e.g.
usr/local/miniconda3/envs/dl/...
(4) build icp so file
(5) mv libicp_op.so ../Icp_op.so
(6) run the code
e.g.
python train.py --logtostderr --checkpoint_dir /output/jikeyun_test/ --data_dir /input/data/ --batch_size 4 --img_height 192 --img_width 320 --icp_weight 0.1 --legacy_mode true
########## compil flownet 2.0 ############################ first try:
(1) error: error: token ""CUDACC_VER is no longer supported
solution: comments /usr/local/cuda/include/crt/common_functions.h line 64: #define CUDACC_VER "CUDACC_VER is no longer supported. Use CUDACC_VER_MAJOR, CUDACC_VER_MINOR, and CUDACC_VER_BUILD instead."
(2) error: core/util/cuda_device_functions.h:32:31: fatal error: cuda/include/cuda.h:
solution: (1) print(tf.sysconfig.get_include())
(2) add /usr/local in CFLAGS
(3) add GPUCFLAGS = --expt-relaxed-constexpr
add -DNDEBUG in CFLAGS
(4) error: libtensorflow_framework.so => not found
Second try:
(1) conda deactivate conda deactivate do not use conda
sampepose/flownet2-tf#28 (comment)
https://blog.csdn.net/davidhopper/article/details/81206673
install cuda 9.0 in /usr/local/cuda-9.0
install cudnn 7.1
intall tensorflow 1.9.0
copy makefile from the link sampepose/flownet2-tf#28 (comment)
(2) error: libtensorflow_framework.so => not found it doesnot matter, we still can run the code
(3) the way to find the libtensorflow_framework.so
TF_LIB=$(python -c 'import tensorflow as tf; print(tf.sysconfig.get_lib())')
TF_INC=$(python -c 'import tensorflow as tf; print(tf.sysconfig.get_include())')
Third try:
(1) download sampepose/flownet2-tf#28 (comment) makefile
(2) remove GLIBCXX_USE_CXX11_ABI=0 in cmakefile !!!
(3) make sure the env: gcc 5.4 g++ 5.4 libtensorflow_framework.so is in the path of python2.7 but not in python 3.6 !!!
(4) download checkpoints in https://drive.google.com/file/d/1cft8EvnsBL5w4-REUeAaVWLeRx39hyHE/view
error: tensorflow 报错 libcusolver.so.8.0: cannot open shared object file: No such file or directory
solution: echo $LD_LIBRARY_PATH /usr/local/cuda-8.0/lib64 sudo ldconfig /usr/local/cuda-8.0/lib64
error: change python3 then can not open terminator
solution: modify /usr/bin/terminator file from /usr/bin/python to /usr/bin/python2
error: AttributeError: module 'tensorflow.estimator' has no attribute 'SessionRunHook'
solution:
(1) Uninstall tensorflow, tensorboard
(2)Uninstall tb-nightly(if it is installed)
(3)Use "pip freeze | grep tensorflow" to check if tensorflow-estimator package has been installed. If so, uninstall it.
(4)Go to site-packages and remove all tensorflow folders related to tensorflow, tensorboard, tensorflow-estimator etc
(5)Reinstall the latest versions of tensorflow and tensorboard
error: tensorflow summary error：tags and values not the same shape [ ] != [16]
solution: tx = tf.reduce_mean(self.egomotion[:, 0, 0], name = 'tx')
tf.summary.scalar('tx' , tx)
instead of tf.summary.scalar('tx' , self.egomotion[:, 0, 0 ])
error: can not use "VideoCapture"
solution: conda install -c conda-forge ffmpeg
conda install -c conda-forge opencv