1. 画面有缺失的躯干，预测骨架是脑补的，一直抖动。
2. 当出现多人的，预测只有一个人能显示，但是预测骨架会在几个人身上来回变。

Define a Keras model for pose classification
print(landmarks.shape)结果(None, 17, 2)

ValueError: Exception encountered when calling layer 'flatten' (type Flatten).

Attempt to convert a value (None) with an unsupported type (<class 'NoneType'>) to a Tensor.

Call arguments received by layer 'flatten' (type Flatten):
• inputs=<KerasTensor: shape=(None, 17, 2) dtype=float32 (created by layer 'tf.math.truediv')>
为什么呀？

可以实现同时识别多个人吗，我测试了下只能识别一个。
要实现同时识别多人，可行吗？求大神指个路
（可能性能要求会比较高？同时识别2个人也成）

林哥你的这个开源项目突然让我想到一个医疗中使用的场景：
医院中有个科室叫做康复科，主要以促进残疾人及患者康复为主要目的。其中不乏很多需要判定患者在治疗中康复运动是否达标或到位，比如一个简单的抬脚、举手、弯腰等动作。
如果有了这个应用，患者或者自己就可以独立完成一些简单的动作训练，从而解放治疗师的时间，服务更多的患者。

看到林哥的样例使用的是摄像头，对于这个输入源，视频按道理也是可以的，毕竟同样使用的都是 SurfaceView ，如果能行，我是想通过选择本地视频，视频中播放患者做康复训练，在绘制出人体部位的同时，同时绘制出某个部位（比如膝盖）在训练中的移动轨迹。

林哥的这个项目确实很棒，从中其实可以挖掘出很多潜在的应用场景，欢迎大家前来讨论。

想要自己训练一些动作，编译大佬的文件报错

报错位置：
converter = tf.lite.TFLiteConverter.from_keras_model(model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
tflite_model = converter.convert()

报错信息：Some of the operators in the model are not supported by the standard TensorFlow Lite runtime. If those are native TensorFlow operators, you might be able to use the extended runtime by passing --enable_select_tf_ops, or by setting target_ops=TFLITE_BUILTINS,SELECT_TF_OPS when calling tf.lite.TFLiteConverter(). Otherwise, if you have a custom implementation for them you can disable this error with --allow_custom_ops, or by setting allow_custom_ops=True when calling tf.lite.TFLiteConverter(). Here is a list of builtin operators you are using: ADD, DIV, EXPAND_DIMS, FLOOR_DIV, FULLY_CONNECTED, GATHER, MAXIMUM, MUL, PACK, REDUCE_MAX, RESHAPE, SOFTMAX, SQRT, SQUEEZE, STRIDED_SLICE, SUB, SUM. Here is a list of operators for which you will need custom implementations: BroadcastTo, Size.

需要增加 这句： converter.allow_custom_ops=True

得到最终：16:23converter = tf.lite.TFLiteConverter.from_keras_model(model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.allow_custom_ops=True
tflite_model = converter.convert()

这样编译通过了 但是得到的tflite的文件运行到AS上面跑不通，想知道up主为什么不需要加就能编译通过

import org.tensorflow.posemon.data.BodyPart这个依赖中的posemon报红该怎么解决 安卓开发不太懂

如果只是手机摄像头的话不够灵活，我一开始想尝试去调用esp32-cam摄像头，安卓技术太菜了，搞了半天没搞出来

希望能增加切换到前置摄像头的支持

设备小米6 类原生Android11

求适配麒麟

难道是不能设置始终允许相机的原因？

在win11安卓子系统虚拟机上监控画面是横着的，希望能增加一个选项调节

上半身动作，增加一个里屏幕太近就提示的警告

如果运用在线下课堂，就可以检测学生上课的时候是不是在看黑板，或者在大学里面通过学生的专注姿态反向评价老师的上课质量。然后每节课下课就生成报告上传到师生的APP上，就是不知道这app做出来会不会喷死，并且刚刚试了一下 一次只能跟踪一个目标。不知道，能否实现同时跟踪多个目标。想做这个项目，大佬能带带吗？

来求语音开源框架

Hi,
I tried loading the movenat_thunder.tflite model with tf.lite.interpreter, using the following code:

import tensorflow as tf
interpreter = tf.lite.Interpreter(model_path='movenat_thunder.tflite')

But was hit with the following error:

ValueError Traceback (most recent call last)
/home/generate_cc_array.ipynb Cell 5 in <cell line: 1>()
----> 1 interpreter = tf.lite.Interpreter(model_path='movenat_thunder.tflite')
2 interpreter.allocate_tensors()
4 input_details = interpreter.get_input_details()[0]

File ~/virtual_environments/utkface/lib/python3.10/site-packages/tensorflow/lite/python/interpreter.py:455, in Interpreter.init(self, model_path, model_content, experimental_delegates, num_threads, experimental_op_resolver_type, experimental_preserve_all_tensors)
448 custom_op_registerers_by_name = [
449 x for x in self._custom_op_registerers if isinstance(x, str)
450 ]
451 custom_op_registerers_by_func = [
452 x for x in self._custom_op_registerers if not isinstance(x, str)
453 ]
454 self._interpreter = (
--> 455 _interpreter_wrapper.CreateWrapperFromFile(
456 model_path, op_resolver_id, custom_op_registerers_by_name,
457 custom_op_registerers_by_func, experimental_preserve_all_tensors))
458 if not self._interpreter:
459 raise ValueError('Failed to open {}'.format(model_path))

ValueError: quantized_dimension must be in range [0, 1). Was 3.Tensor 33 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 36 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 40 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 44 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 48 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 52 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 56 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 60 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 64 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 68 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 72 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 76 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 80 has invalid quantization parameters.quantized_dimension must be in range [0, 1). Was 3.Tensor 84 has invalid quantization parameters.

Tensorflow version : 2.9.1
Python version: 3.10.4

How to solve this issue? Please guide

Thanks

不把NNAPI设为默认项就可以了。

机型：三星Galaxy S9，打开会崩溃闪退，AS报错信息：Process: lyi.linyi.posemon, PID: 31265
java.lang.IllegalArgumentException: Internal error: Failed to apply delegate: NN API returned error ANEURALNETWORKS_OP_FAILED at line 4274 while completing NNAPI compilation.

Node number 157 (TfLiteNnapiDelegate) failed to prepare.

切换后

切换前

888应该是支持NNAPI的吧

如題～
測試機型：1 III、10 II、XA2、XZ2。以上四機型皆在取得相機權限後閃退。

from data import BodyPart
from ml import Movenet
where are those two modules ?

三星note 10+ 手机 不能运行

AttributeError Traceback (most recent call last)
in
12 image_bad = tf.io.decode_jpeg(image_bad)
13 person = detect(image_bad)
---> 14 _ = draw_prediction_on_image(image_bad.numpy(), person, crop_region=None,
15 close_figure=False, keep_input_size=True)

in draw_prediction_on_image(image, person, crop_region, close_figure, keep_input_size)
18 """
19 # Draw the detection result on top of the image.
---> 20 image_np = utils.visualize(image, [person])
21
22 # Plot the image with detection results.

AttributeError: module 'utils' has no attribute 'visualize'
类似这样的好多地方都找不到我要如何进行操作呢？

Hello，用华为mate 30可以正常运行APP，但是相机取景框焦距恒定，需要把手机放在离人很远的地方（不确定其他手机是否也是这样）。对于空间比较小的房间几乎无法使用，想问一下是否可以增加调节相机焦距的功能？

电脑没装环境

打开软件之后,并没有打开摄像头,什么反应都没有,请问是需要自己编译才可以吗?

啊?
tested on human mate

作者可以分一个监测小学生写字低头提醒的版本吗？叫爸爸康康。

在第一次提醒之后，依然保持动作不变，后续不会再提醒