测试的时候，有时候会出现人脸上有好几框的情况，如何解决？

我将landmark用的数据集换成了我自己的数据集，process对数据的处理都没有问题，也成功生成了tfcord文件，但是在训练p-net的时候却遇到了如下错误。求解。
2019-08-03 17:03:27.010008: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
Traceback (most recent call last):
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1322, in _do_call
return fn(*args)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1307, in _run_fn
options, feed_dict, fetch_list, target_list, run_metadata)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1409, in _call_tf_sessionrun
run_metadata)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Reshape cannot infer the missing input size for an empty tensor unless all specified input sizes are non-zero
[[Node: gradients/PNet/TopKV2_2_grad/Reshape = Reshape[T=DT_INT32, Tshape=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"](PNet/TopKV2_2:1, gradients/PNet/TopKV2_2_grad/stack)]]

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "train.py", line 58, in
main(parse_arguments(sys.argv[1:]))
File "train.py", line 41, in main
train(net_factory,prefix,end_epoch,base_dir,display,lr)
File "/home/lishuwei/dsx/tensorflow-MTCNN-master/train/train_model.py", line 122, in train
,,summary = sess.run([train_op, lr_op ,summary_op], feed_dict={input_image: image_batch_array, label: label_batch_array, bbox_target: bbox_batch_array,landmark_target:landmark_batch_array})
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 900, in run
run_metadata_ptr)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1135, in _run
feed_dict_tensor, options, run_metadata)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1316, in _do_run
run_metadata)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1335, in _do_call
raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Reshape cannot infer the missing input size for an empty tensor unless all specified input sizes are non-zero
[[Node: gradients/PNet/TopKV2_2_grad/Reshape = Reshape[T=DT_INT32, Tshape=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"](PNet/TopKV2_2:1, gradients/PNet/TopKV2_2_grad/stack)]]

Caused by op 'gradients/PNet/TopKV2_2_grad/Reshape', defined at:
File "train.py", line 58, in
main(parse_arguments(sys.argv[1:]))
File "train.py", line 41, in main
train(net_factory,prefix,end_epoch,base_dir,display,lr)
File "/home/lishuwei/dsx/tensorflow-MTCNN-master/train/train_model.py", line 82, in train
train_op,lr_op=optimize(base_lr,total_loss_op,num)
File "/home/lishuwei/dsx/tensorflow-MTCNN-master/train/train_model.py", line 159, in optimize
train_op = optimizer.minimize(loss, global_step)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/training/optimizer.py", line 414, in minimize
grad_loss=grad_loss)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/training/optimizer.py", line 526, in compute_gradients
colocate_gradients_with_ops=colocate_gradients_with_ops)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py", line 494, in gradients
gate_gradients, aggregation_method, stop_gradients)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py", line 636, in _GradientsHelper
lambda: grad_fn(op, *out_grads))
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py", line 385, in _MaybeCompile
return grad_fn() # Exit early
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py", line 636, in
lambda: grad_fn(op, *out_grads))
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/nn_grad.py", line 978, in _TopKGrad
ind_2d = array_ops.reshape(op.outputs[1], array_ops.stack([-1, ind_lastdim]))
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/gen_array_ops.py", line 6113, in reshape
"Reshape", tensor=tensor, shape=shape, name=name)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3392, in create_op
op_def=op_def)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1718, in init
self._traceback = self._graph._extract_stack() # pylint: disable=protected-access

...which was originally created as op 'PNet/TopKV2_2', defined at:
File "train.py", line 58, in
main(parse_arguments(sys.argv[1:]))
[elided 0 identical lines from previous traceback]
File "train.py", line 41, in main
train(net_factory,prefix,end_epoch,base_dir,display,lr)
File "/home/lishuwei/dsx/tensorflow-MTCNN-master/train/train_model.py", line 80, in train
label,bbox_target,landmark_target,training=True)
File "/home/lishuwei/dsx/tensorflow-MTCNN-master/train/model.py", line 41, in P_Net
landmark_loss=landmark_ohem(landmark_pred,landmark_target,label)
File "/home/lishuwei/dsx/tensorflow-MTCNN-master/train/model.py", line 219, in landmark_ohem
square_error,_=tf.nn.top_k(square_error,k=keep_num)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 2352, in top_k
return gen_nn_ops.top_kv2(input, k=k, sorted=sorted, name=name)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 7660, in top_kv2
"TopKV2", input=input, k=k, sorted=sorted, name=name)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3392, in create_op
op_def=op_def)
File "/home/lishuwei/anaconda3/envs/dsx_tf/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1718, in init
self._traceback = self._graph._extract_stack() # pylint: disable=protected-access

InvalidArgumentError (see above for traceback): Reshape cannot infer the missing input size for an empty tensor unless all specified input sizes are non-zero
[[Node: gradients/PNet/TopKV2_2_grad/Reshape = Reshape[T=DT_INT32, Tshape=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"](PNet/TopKV2_2:1, gradients/PNet/TopKV2_2_grad/stack)]]

请问大佬这个该如何解决？

i want to ask how to set the parameter to train data by using gpu?

你好哦！我想问一下r_net的neg、pos、part的数据集只是用p_net进行hard_example处理嘛？需不需要还对原始的wideface数据集进行处理？

在其它博客看到如下的描述：
由原始图片和PNet生成预测的bounding boxes；
输入原始图片和PNet生成的bounding box，通过RNet，生成校正后的bounding box；
输入元素图片和RNet生成的bounding box，通过ONet，生成校正后的bounding box和人脸面部轮廓关键点。

请问boundbox_regression输出的四个值，是关于【左上角坐标，宽，高】的，还是【左上角坐标，右下角坐标】的？代码貌似和论文不一致，求教。

感谢您的分享，这个目前最顺利的tf mtcnn 训练和测试项目。

我想请教data目录下这个图像路径文件生成的脚本在哪？

hi, 请问下NMS是用于测试还是训练的？

作者你好，我想问下有关训练标签的问题。我自己制作的小的数据集已经标记完成，也已经做好对应的txt文件。

但是看到在data文件下wider face的训练标签文件有两个。

一个是wider_face_train.txt,一个是wider_face_train_bbx_gt.txt，train_bbx_gt文件的边框格式是x1,y1,w,h,我对应的看了下train.txt文件，发现这个格式应该是x1,y1,x2,y2的格式，但是不明白为啥边框数据还带着小数，虽然这个对边框位置感觉影响不大

另一个问题是根据您仓库中gen_12net_data.py中这一句来看
，真正的训练标签格式是x1,y1,x2,y2格式。请问您这样理解是否是对的

含有人脸的图像输出，最后头像上有好几个框，不止一个框，只是为什么

你好，p网络检测出来的框，在输入R网络前，没有进行回归吗？

想问问为什么更新了picture之后，进行test之后output是没有更新的。
希望可以得到您的解答

请问楼主，你在util函数中，convert_to_square子函数转换为更大正方形极有可能超出图像索引，请问这样是否正确？

可以自己训练一个其他的模型吗？比如：检测猫头或狗头😁
如果可以，代码部分有要改的吗？谢谢

train.py: error: the following arguments are required: input_size
请问这个报错是什么原因

你好，小白有一个脑残问题，既然lfw数据集既有bbox数据，也有landmark数据，为什么要分别用wider_face数据集来训练bbox，lfw来训练landmark呢？ 而且这样就造成了P-net的训练过程中每条样本就只能计算一种损失，期待您的回复

运行 test.py
只输出一个结果图片并一直保持不变，
这是为什呢？
并有如下提示：
**Qt: Cannot set locale modifiers:**

请问作者知道哪里有mtcnn的预训练模型吗？

你好，请问这个模型训练出来在FDDB数据上的精度能达到多少呢？

请问除了图片，摄像头的检测怎么使用？十分感谢希望能够解答

请问一下，产生P-net的训练数据时。所有的postive,part的偏移量值是否都在-1到1之间。而landmark的偏移量值是否都在0到1之间？

您好，不知道该MTCNN项目能不能在windows上运行呢？如不能的话，有windows上相关的资源吗？如您方便时能回复，万分感谢

经过PNet检测后输出的图像是不是明显变少了呀？
这些输出图像用来训练RNet，但是我发现这些图像的3类数据特别少
比如我用来训练PNet的数据多达1750000
但是用来训练RNet的数据只有7000，pos只有63个。。。
请问博主有发生这种情况吗？该怎么解决呢？

使用conda install tensorflow-gpu==1.8.0安装，cuda=9.0 cudnn=7.6.5，并使用os.environ["CUDA_VISIBLE_DEVICES"] = "0"
指定gpu，但却无法使用gpu训练。

Hi, I'm trying this project. The speed on GPU is lower than CPU in training phase, WHY?

我的pnet训练准确率为93%左右，在图片较大时推断时会生成成千上万个候选框，导致NMS就会花费2s左右。且网络推断速度很慢pnet0.5s左右，gpu是1080ti，已绝望，请问楼主的推断速度是多少呢，有没有注意过候选框数量的问题

请问可以修改为人头检测吗，除了去掉landmark，修改训练集还需要修改哪里吗

你好，非常高兴能够看到这么通俗易懂的代码。
因为项目需要，我对代码进行了修改，删除了landmark识别，将网络的保存格式转为了pb模型，并用了tf.layers api进行了改写网络，加入了batch_nromal。

目前，程序运行至生成hard_example 12 阶段。
前面几张图片处理速度还比较快，大概3秒一张，但是处理到800多张的时候，却需要2分钟一张了，想问一下，有没有人遇到这个问题，可以一起讨论一下。

centos7
8 core
16 mem
1080ti

./train/models.py 166

在求face classification的时候，代码中loss中使用的是

P1=y*log(p(y))

（这个应该是用在多分类问题中吧），但是论文里面说了这个是二分类问题，应该选用的是

P2=y*log(p(y))+(1-y)*log(1-p(y))

毕竟这个两个求出的相关交叉熵函数是不一样的，可能在optimizer的时候有差异。
例如: y =1,p(y)=0.0, P1=log(0.6); P2=log(0.6)+log(0.4)

（优化未能充分说明，结果待验证）

楼主怎么看？

lunar@lunar-virtual-machine:~/PycharmProjects/tensorflow_mtcnn/preprocess$ python gen_hard_example.py 24
2019-05-12 09:33:06.690596: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
载入数据
8247it [2:47:10, 1.26it/s]Traceback (most recent call last):
File "gen_hard_example.py", line 193, in
main(parse_arguments(sys.argv[1:]))
File "gen_hard_example.py", line 70, in main
detectors,_=mtcnn_detector.detect_face(test_data)
File "../detection/MtcnnDetector.py", line 87, in detect_face
boxes, boxes_c, landmark = self.detect_rnet(im, boxes_c)
File "../detection/MtcnnDetector.py", line 181, in detect_rnet
cls_scores, reg, _ = self.rnet_detector.predict(cropped_ims)
File "../detection/detector.py", line 63, in predict
return np.concatenate(cls_prob_list, axis=0), np.concatenate(bbox_pred_list, axis=0), np.concatenate(landmark_pred_list, axis=0)
ValueError: need at least one array to concatenate

请问一下这是什么报错？谢谢。

您好，跑到第二个步骤
2 python gen_landmark_aug.py 12 生成pnet的landmark数据，

文件中只有python gen_landmark_aug.py 这个文件，请问要在哪里给input_size赋值？
问题可能比较小白～期待您的答复。

`image_batch_array, label_batch_array, bbox_batch_array, landmark_batch_array = sess.run([image_batch, label_batch, bbox_batch, landmark_batch])

随机翻转图像

image_batch_array, landmark_batch_array = random_flip_images(image_batch_array, label_batch_array, landmark_batch_array)
_, _, summary = sess.run([train_op, lr_op, summary_op], feed_dict={input_image: image_batch_array,label: label_batch_array,bbox_target: bbox_batch_array,landmark_target: landmark_batch_array})`