The tensorflow CI is probably failing because of a new tensorflow version. Sadly the Industrial CI doesn't show the installed version of pip packages.

Update: 2.12 does work, 2.13 fails

Remove pytorch
Add age_gender

Or at least announce it somewhere?

Also the flag says: travis failing

@reinzor, We now use the ROS buildfarm right?

Hey,

If you label some images using the annotation tool, have underscores in the names of the labels and use rosrun tensorflow_ros retrain ..., the underscores will be replaced with spaces in output_lables.txt.
If you now use the object_recognition_node to test some new images, you can't just merge the unverified and verified folders afterwards to generate a new and larger training set, because the names differ (underscores vs spaces). The spaces in the labels also could lead to some other problems.

I don't know, if this is the intended behavior (because in your example underscores are used).

Inserting an underscore in line 263 in tf_retrain.py will fix this problem.
Old: label_name = re.sub(r'[^a-z0-9]+', ' ', dir_name.lower())
New: label_name = re.sub(r'[^a-z0-9_]+', ' ', dir_name.lower())

Labels should not contain spaces, but underscores for compatibility in nl_robot_console.
One example now is the label "ice tea".

You get a dropdown now for the recognitions but this is not sorted or highlighted based on the recognition probability.

I was trying out this package for the first time on Ubuntu 16.04, ROS Kinetic and Tensorflow 1.8.0. First I annotated two objects with 4-5 images each using the annotation_gui. Then I ran tensorflow_ros_rqt, pointed to my image directory, and clicked the Train button but received the attached popup error window.

Not sure how to proceed from here.

Thanks!

Should be python-tensorflow-pip https://github.com/ros/rosdistro/pull/13724/files

['Traceback (most recent call last):\n', ' File "/opt/ros/kinetic/lib/python2.7/dist-packages/rospy/impl/tcpros_service.py", line 625, in _handle_request\n response = convert_return_to_response(self.handler(request), self.response_class)\n', ' File "/home/amigo/ros/kinetic/system/src/tensorflow_ros/scripts/object_recognition_node", line 73, in _recognize_srv_callback\n if predictions:\n', 'ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()\n']

Hi again .

the .pb and .txt file created successfully by train node #58

now I use object_recognition_node but nothing created for me
what should I do ?

and my second question is , it is possible to use this .pb file in object detection package?
https://github.com/cagbal/ros_people_object_detection_tensorflow
for example in this package ?? this package use Model Zoo of tensorflow object detection and I need to replace my own model . is it possible ? if yes please explain it

There is this catkin_make error:
CMake Error at image_recognition/image_recognition_analysis/CMakeLists.txt:14 (message):
catkin_lint failed

model = YOLO("/some/path/to/an/yolo/model/yolov8n-pose.pt")

This does work and controls the download location, so it isn't downloaded in the current working directory.

Message: Error processing request: ('Failed to run tensorflow session: %s', ValueError("Cannot feed value of shape () for Tensor u'Cast:0', which has shape '(?, ?, 3)'",))
['Traceback (most recent call last):\n', ' File "/opt/ros/kinetic/lib/python2.7/dist-packages/rospy/impl/tcpros_service.py", line 625, in _handle_request\n response = convert_return_to_response(self.handler(request), self.response_class)\n', ' File "/home/amigo/ros/kinetic/system/src/tensorflow_ros/scripts/object_recognition_node", line 61, in _recognize_srv_callback\n sorted_result = self._recognizer.classify(self._bgr_image)\n', ' File "/home/amigo/ros/kinetic/system/src/tensorflow_ros/src/tensorflow_ros/object_recognizer.py", line 59, in classify\n raise Exception("Failed to run tensorflow session: %s", e)\n', 'Exception: ('Failed to run tensorflow session: %s', ValueError("Cannot feed value of shape () for Tensor u'Cast:0', which has shape '(?, ?, 3)'",))\n']

Thank you for your good work.

Is it correct that the ros package: openface_ros from interaction_lab is based on a completely different openface library (https://github.com/interaction-lab/openface_ros)?

As far as I understand it, there exists one openface library from the cambridge university and the one your package is based on from the cmu university. Is that right?

The package one can download using: sudo apt-get install ros-kinetic-image-recognition is based on your package right?

Best,
Max

See if the web API is down just temporarily or definitively

For several RoboCup challenges, we have to deal with people and their properties.
In order to do this, the idea is to associate faces (from OpenFace) and skeletons (from OpenPose). The association can be done by assigning the skeleton who's 'nose' joint is in the bounding box of a face;s bounding box.

Then, based on that pair, we can assign other properties, such as:

• Name
• Age
• Gender
• posture: (list of string which can be 'waving', 'holding_drink', 'sitting' etc.)
• Has_beard, has_glasses
• Length (is_tall, is_short etc)

First step is to do this merging and then put these in ED through it;s ROS service API

[ERROR] [1513100614.242175]: ~save_images_folder is not defined but we would like to save images ...
Traceback (most recent call last):
File "/home/amigo/ros/kinetic/system/src/openface_ros/scripts/face_recognition_node", line 204, in
service_publish_result_image)
File "/home/amigo/ros/kinetic/system/src/openface_ros/scripts/face_recognition_node", line 50, in init
rospy.signal_shutdown()
TypeError: signal_shutdown() takes exactly 1 argument (0 given)

Hi,

I'm trying to compile the image_recognition_jetson package. After cloning and compiling the jetson-inference package as done in install_jetson_inference.bash, Cmake can't (fully) find the package.

So when I build the image_recognition_jetson package I get the following warnings:

CMake Warning at image_recognition/image_recognition_jetson/CMakeLists.txt:25 (find_package):
  Found package configuration file:

    /usr/local/share/jetson-inference/cmake/jetson-inferenceConfig.cmake

  but it set jetson-inference_FOUND to FALSE so package "jetson-inference" is
  considered to be NOT FOUND.  Reason given by package:

  The following imported targets are referenced, but are missing:
  jetson-utils



CMake Warning at image_recognition/image_recognition_jetson/CMakeLists.txt:35 (message):
  Jetson-inference not found, using a mock instead ...


CMake Warning at image_recognition/image_recognition_jetson/CMakeLists.txt:36 (message):
  The library can be installed using scripts/install_jetson_inference.bash

Has anyone experienced anything similar before? Any help would be appreciated.

System info:
Jetson tx2
Ubuntu 16.04
ros kinetic

Hey,
Thanks for this nice library.

I was playing around with the tensorflow_ros package and was confused that my red ball was recognized as a blue ball and vice versa.
Looked into the issue and found out that evaluate_classifier, get_object_recognition and object_recognition_node all send BGR images to object_recognizer.py, but that script expects the images to be in RGB format.
The easiest solution is to transform the images to RGB in object_recognizer.py before they get send to tensorflow.

Best regards,
Jannis

W tensorflow/core/framework/op_def_util.cc:332] Op BatchNormWithGlobalNormalization is deprecated. It will cease to work in GraphDef version 9. Use tf.nn.batch_normalization().

Hi . First at all apologize for my bad language

when I try to train my image folder an error occurs

Something went wrong during retraining, 'Unknown command line flag 's''

would you please help me .
I guess , maybe I have mistake in annotation part

It's annotated folder that , image_recognition_rqt make for me . is it true or no? if no what can I do for correct it?

Traceback (most recent call last):
File "/home/amigo/ros/kinetic/system/src/tensorflow_ros/scripts/evaluate_classifier", line 42, in
classification_score_matrix.write_to_file(args.output)
File "/home/amigo/ros/kinetic/system/src/image_recognition_util/src/image_recognition_util/classification_score_matrix.py", line 101, in write_to_file
row = dict({'ground_truth': ground_truth}, **dict(scores))
TypeError: cannot convert dictionary update sequence element #0 to a sequence

CI fails on installing a new version, 1.3.1, of SciPy which requires Python 3.5. Scipy 1.2.X is still compatible with Python 2.7.

Discussion can be found here #33

Or move the bodypart messages there

rosrun tensorflow_ros object_recognition_node**.py** _graph_path:=/home/amigo/MEGA/data/robotics_testlabs/training_data_Josja/models_15/output_graph.pb _labels_path:=/home/amigo/MEGA/data/robotics_testlabs/training_data_Josja/models_15/output_labels.txt

We can use the diagnostics to check if the Jetson board that runs OpenPose is performing OK or is even connected at all.

[ERROR] [1509800338.863343]: [32FC1] is not a color format. but [bgr8] is. The conversion does not make sense
[ERROR] [1509800338.863639]: bad callback: <bound method TestPlugin._image_callback of <image_recognition_rqt.test.TestPlugin object at 0x7ff759ead218>>
Traceback (most recent call last):
File "/opt/ros/kinetic/lib/python2.7/dist-packages/rospy/topics.py", line 750, in _invoke_callback
cb(msg)
File "/home/amigo/ros/kinetic/system/src/image_recognition_rqt/src/image_recognition_rqt/test.py", line 134, in _image_callback
self._image_widget.set_image(cv_image)
UnboundLocalError: local variable 'cv_image' referenced before assignment

http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz

Openpose is known to consume a lot of resources when run continuously.

Always having an active image subscriber/publisher even though the results are needed at low frequency drains out the battery because of exhaustive GPU use.

A proposed solution is to have additional parameter in the node that starts the node in either service only mode, or both in service and topic mode.

Something went wrong during retraining, ''module' object has no attribute 'scalar_summary''

1.11 is not working, they transitions to the tensorflow hub

I would like to add a package image_recognition_tests that test several image batches and show regression of these tests in CI.

Motivation: the output from for example openpose does not fit in the Recognition message definition ..

We'd like to standardize all machine learning applications at Tech
Unitd@Home on PyTorch.

A start has been made in #146

Preferably use the same models, but interface it via PyTorch rather than Keras/Tensorflow/Caffe etc.