在 Ubuntu20.02下编译 FaceRecognition_MTCNN_FaceNet
1. compile FaceRecognition_MTCNN_FaceNet-master
git clone https://github.com/Chanstk/FaceRecognition_MTCNN_FaceNet.git
cd FaceRecognition_MTCNN_FaceNet
mkdir build && cd build
cmake .. && make +----------------------------solve compile error --------------------+
报错信息 : error adding symbols: DSO missing from command line
原因 : shared lib called by main calls another shared lib, but CMakeLists.txt has not add -llibother.so
解决 : CXX_FLAGS += -Wl,--copy-dt-needed-entries
+----------------------------solve compile error --------------------+
报错信息 : /usr/bin/ld: warning: libmkl_intel_lp64.so.2, needed by /usr/local/lib/libopencv_core.so.4.8.0, not found (try using -rpath or -rpath-link)
解决 : 编辑 CMakeLists.txt，增加
link_directories( /opt/intel/oneapi/mkl/2023.2.0/lib/intel64 )
+----------------------------solve compile error --------------------+
报错信息 : 2023-09-08 14:13:58.777362: E /home/rd/NN/FaceRecognition_MTCNN_FaceNet-master/src/main.cpp:107] Read proto
解决 : 编辑 src/main.cpp
- string graph_path = "./model/20170323-142841.pb";
+ string graph_path = "./model/mtcnn_frozen_model.pb";
+--------------------------------------------------------------------------------------------------------------------+
报错信息 :
has not enablesd AVX2 FMA
2023-09-08 14:16:17.991925: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used.
2023-09-08 14:16:18.070278: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-09-08 14:16:18.218639: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:375] MLIR V1 optimization pass is not enabled
terminate called after throwing an instance of 'cv::Exception'
what(): OpenCV(4.8.0) /home/rd/opencv/opencv-4.8.0/modules/core/src/arithm.cpp:652: error: (-215:Assertion failed) type2 == CV_64F && (sz2.height == 1 || sz2.height == 4) in function 'arithm_op'
解决方法 :
mkdir model;
cp ../../facenet-compare-cpp/facenet-compare-cpp-master/models/mtcnn_frozen_model.pb model/
+--------------------------------------------------------------------------------------------------------------------+
(https://www.intel.com/content/www/us/en/developer/articles/guide/optimization-for-tensorflow-installation-guide.html)
wget https://repo.anaconda.com/archive/Anaconda3-2023.07-2-Linux-x86_64.sh
sudo chmod +x Anaconda3-2023.07-2-Linux-x86_64.sh
sudo ./Anaconda3-2023.07-2-Linux-x86_64.sh
sudo conda install intel-tensorflow -c intel [ Fail ] bazel build --cxxopt=-D\_GLIBCXX\_USE\_CXX11\_ABI=0 --copt=-march=corei7-avx --copt=-mtune=core-avx-i --copt=-O3 --copt=-Wformat --copt=-Wformat-security --copt=-fstack-protector --copt=-fPIC --copt=-fpic --linkopt=-znoexecstack --linkopt=-zrelro --linkopt=-znow --linkopt=-fstack-protector //tensorflow/tools/pip_package:build_pip_package
[ OK ] bazel clean
bazel build -c opt --copt=-msse3 --copt=-msse4.1 --copt=-msse4.2 --copt=-mavx --copt=-mavx2 --copt=-mfma //tensorflow/tools/pip_package:build_pip_package
2. 运行 facenet 的测试脚本
git clone https://github.com/davidsandberg/facenet.git
2.1 图像对齐
for N in {1..4}; do \
python3 src/align/align_dataset_mtcnn.py \
~/datasets/lfw/raw \
~/datasets/lfw/lfw_mtcnnpy_160 \
--image_size 160 \
--margin 32 \
--random_order \
--gpu_memory_fraction 0.25 \
& done (https://github.com/davidsandberg/facenet/wiki/Validate-on-lfw)
+--------------------------------------------------solve compile error ----------------------------------------------+
报错信息 : AttributeError: module ‘tensorflow‘ has no attribute 'GPUOptions'
AttributeError: module 'tensorflow' has no attribute 'Session'
AttributeError: module 'tensorflow' has no attribute 'variable_scope'
cause : Tensorflow 1.X和 2.X不兼容。
Solution : [-]sed -i "s/tf.GPUOptions/tf.compat.v1.GPUOptions/g" */*.py */*/*.py */*/*/*.py
[-]sed -i "s/tf.Session/tf.compat.v1.Session/g" */*.py */*/*.py */*/*/*.py
[-]sed -i "s/tf.ConfigProto/tf.compat.v1.ConfigProto/g" */*.py */*/*.py */*/*/*.py
[+]sed -i "s/import tensorflow as tf/import tensorflow.compat.v1 as tf/g" */*.py */*/*.py */*/*/*.py
+--------------------------------------------------solve compile error ----------------------------------------------+
报错信息 : ValueError: Object arrays cannot be loaded when allow_pickle=False
cause : Tensorflow 1.X和 2.X不兼容。
Solution : 1. pip install numpy=1.16.2
2. vim src/align/detect_face.py +85
- data_dict = np.load(data_path, encoding='latin1' ).item() #pylint: disable=no-member
+ data_dict = np.load(data_path, encoding='latin1', allow_pickle=True).item() #pylint: disable=no-member
+--------------------------------------------------solve compile error ----------------------------------------------+
报错信息 : AtributeError: 'int' object has no attribute 'value'
原因 : Tensorflow 1.X和 2.X不兼容。
解决方法 : 编辑 src/align/detect_face.py +194
- feed_in, dim = (inp, input_shape[-1].value)
+ feed_in, dim = (inp, input_shape[-1]) +--------------------------------------------------solve compile error ----------------------------------------------+
报错信息 :
AttributeError: scipy.misc is deprecated and has no attribute imread
AttributeError: scipy.misc is deprecated and has no attribute imresize.
原因 : 官方scipy中提到，imread is deprecated! imread is deprecated in SciPy 1.0.0, and will be removed in 1.2.0. Use imageio.imread instead
解决方法 : sudo pip3 install imageio
sed -i "s/misc.im/imageio.im/g" */*.py */*/*.py
sed -i "s/from scipy import misc/import imageio.v2 as imageio/g" */*.py */*/*.py */*/*/*.py
vim src/align/align_dataset_mtcnn.py
+ import imageio
- img = misc.imread(image_path)
+ img = imageio.imread(image_path)
+--------------------------------------------------solve compile error ----------------------------------------------+
报错信息 : AttributeError: module 'imageio.v2' has no attribute 'imresize'
原因 : Tensorflow 1.X和 2.X不兼容。
解决方法 :
sed -i "s/imageio.imresize(cropped,/Image.fromarray(cropped).resize(/g" */*.py */*/*.py
sed -i "s/imageio.imresize(img,/Image.fromarray(img).resize(/g" */*.py */*/*.py */*/*/*.py
sed -i "s/import imageio.v2 as imageio/import imageio.v2 as imageio\nfrom PIL import Image/g" */*.py */*/*.py */*/*/*.py
sed -i "s/interp='bilinear'/Image.BILINEAR/g" */*.py */*/*.py +--------------------------------------------------solve compile error ----------------------------------------------+
# sudo apt install libjpeg-dev libtiff-dev
# sudo pip install imageio iio # image I/O
+--------------------------------------------------solve compile error ----------------------------------------------+
报错信息 : div (from tensorflow.python.ops.math_ops) is deprecated and will be removed
解决方法 : https://docs.w3cub.com/tensorflow~1.15/div.html
1. 编辑 src/align/detect_face.py : 214 行
- tf.div
+ tf.compat.v1.div\
2.2 validate on lfw dataset
python src/validate_on_lfw.py \
~/datasets/lfw/lfw_mtcnnpy_160 \
~/models/facenet/20180402-114759 \
--distance_metric 1 \
--use_flipped_images \
--subtract_mean \
--use_fixed_image_standardization
+--------------------------------------------------solve compile error ----------------------------------------------+
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions.
WARNING:tensorflow:From /home/rd/NN/facenet/src/facenet.py:112: py_func (from tensorflow.python.ops.script_ops) is deprecated and will be removed in a future version.
+--------------------------------------------------solve compile error ----------------------------------------------+
problem : WARNING:tensorflow:From /home/rd/NN/facenet/src/facenet.py:112: py_func (from tensorflow.python.ops.script_ops) is deprecated and will be removed in a future version.
Instructions for updating:
tf.py_func is deprecated in TF V2. Instead, there are two
options available in V2.
- tf.py_function takes a python function which manipulates tf eager
tensors instead of numpy arrays. It's easy to convert a tf eager tensor to
an ndarray (just call tensor.numpy()) but having access to eager tensors
means `tf.py_function`s can use accelerators such as GPUs as well as
being differentiable using a gradient tape.
- tf.numpy_function maintains the semantics of the deprecated tf.py_func
(it is not differentiable, and manipulates numpy arrays). It drops the
stateful argument making all functions stateful.
Solution : vim src/facenet.py +112
+--------------------------------------------------solve compile error ----------------------------------------------+
problem : WARNING:tensorflow:From /home/rd/NN/facenet/src/facenet.py:131: batch_join (from tensorflow.python.training.input) is deprecated and will be removed in a future version.
Instructions for updating:
Queue-based input pipelines have been replaced by `tf.data`. Use `tf.data.Dataset.interleave(...).batch(batch_size)` (or `padded_batch(...)` if `dynamic_pad=True`).
+--------------------------------------------------solve compile error ----------------------------------------------+
problem : WARNING:tensorflow:From /home/rd/.local/lib/python3.8/site-packages/tensorflow/python/training/input.py:738: QueueRunner.__init__ (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
WARNING:tensorflow:From /home/rd/.local/lib/python3.8/site-packages/tensorflow/python/training/input.py:738: add_queue_runner (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
sudo pip install tf-slim
sed -i "s/import tensorflow.contrib.slim as slim/import tf_slim as slim/g" */*.py */*/*.py
2.3 比较图像
python3 src/compare.py ~/NN/facenet-pre-trained-model/20180408-102900/20180408-102900.pb \
~/datasets/lfw/Tony_Liu/Tony_Liu_0001.jpg \
~/datasets/lfw/lfw_mtcnnpy_160/Tony_Blair/Tony_Blair_0001.png \
~/opencv/study/video/Face-Samples/stark-face/stark-6.png \
~/opencv/study/video/Face-Samples/stark-face/stark-2.png \
~/opencv/study/video/Face-Samples/tony-face/tony-2.png \
~/opencv/study/video/Face-Samples/tony-face/tony-5.png +--------------------------------------------------solve compile error ----------------------------------------------+
problem : raise ValueError("Expect x to not have duplicates")
ValueError: Expect x to not have duplicates
ERROR: Could not find a version that satisfies the requirement numpy<1.23.0,>=1.16.5 (from scipy==1.7.1) (from versions: none)
ERROR: No matching distribution found for numpy<1.23.0,>=1.16.5 (from scipy==1.7.1)
cause : scipy包下得interpolate.interp1d()函数问题
requirement : numpy<1.23.0,>=1.16.5 (from scipy==1.7.1)
solution ：降低scipy版本，我的是scipy（1.10.1）版本现换为版本1.7.1，可行！
pip uninstall numpy scipy
pip install scipy==1.7.1
+--------------------------------------------------solve compile error ----------------------------------------------+
problem : ValueError: Node 'gradients/InceptionResnetV1/Bottleneck/BatchNorm/cond/FusedBatchNorm_1_grad/FusedBatchNormGrad' has an _output_shapes attribute inconsistent with the GraphDef for output #3: Dimension 0 in both shapes must be equal, but are 0 and 512. Shapes are [0] and [512]
cause : refer to https://github.com/davidsandberg/facenet/issues/1227 / https://github.com/openvinotoolkit/openvino/pull/11078
Solution : For those who undergo this problem. I would suggest following actions:
Add the directive "import tensorflow.compat.v1 as tf" to the corresponding .py files.
Use the arguments to specify the model file and pair.txt with absolute full path, as following,
python3 FaceNet/src/validate_on_lfw.py ../Inventory/Aligned /Users/xxxx/Projects/Inventory/Models/20180402-114759.pb --distance_metric 1 --use_flipped_images --subtract_mean --use_fixed_image_standardization --lfw_pairs /Users/xxxx/Projects//FaceNet/data/pairs.txt
+--------------------------------------------------solve compile error ----------------------------------------------+
problem : 2023-09-13 17:14:46.844967: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:354] MLIR V1 optimization pass is not enabled
Traceback (most recent call last):
File "src/compare.py", line 131, in
main(parse_arguments(sys.argv[1:]))
File "src/compare.py", line 42, in main
images = load_and_align_data(args.image_files, args.image_size, args.margin, args.gpu_memory_fraction)
File "src/compare.py", line 111, in load_and_align_data
prewhitened = facenet.prewhiten(aligned)
File "/home/rd/NN/facenet/src/facenet.py", line 220, in prewhiten
y = np.multiply(np.subtract(x, mean), 1/std_adj)
ValueError: operands could not be broadcast together with shapes (160,160,3) (2,)
solution : vim src/facenet.py +220
std_adj = np.maximum(std, 1.0/np.sqrt(np.prod(x.size)))