run detect_*.py errors: Invalid argument: NodeDef mentions attr 'T' not in Op<name=Where about handtracking

[gangyahaidao]

thank you for your reply, i run this cmd:
python3 -c 'import tensorflow as tf; print(tf.__version__)'
output: 1.4.0
maybe it's version mismatch, i will try to fix it and show the result

[namheegordonkim]

You can copy it to the current directory and use it as follows

No you can't! Like @TyrionZK mentions you need to install Tensorflow object detection API. Do you have instructions for doing so?

[victordibia]

This error looks like the same in issue #1 .
It appears to be an error associated with different tensorflow versions.
The solution is to generate your own frozen graph from the model checkpoint I provide.

See issue #1 for more details.

-V.

[gangyahaidao]

after i reinstall tensor-flow v1.4.0-rc0, it works fine! thanks@victordibia
but i got such errors, when startup comes up CUDA_ERROR_OUT_OF_MEMORY

`rosrobot@rosrobot:~/git/handtracking$ python3 detect_multi_threaded.py
{'num_hands_detect': 2, 'score_thresh': 0.2, 'im_height': 180.0, 'im_width': 320.0} Namespace(display=1, fps=1, height=200, num_hands=2, num_workers=4, queue_size=5, video_source=0, width=300)
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
2018-03-28 16:09:03.123319: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-28 16:09:03.123576: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-28 16:09:03.123690: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-28 16:09:03.124384: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.55GiB
2018-03-28 16:09:03.124405: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-03-28 16:09:03.124661: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.55GiB
2018-03-28 16:09:03.124677: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-03-28 16:09:03.124862: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.55GiB
2018-03-28 16:09:03.124896: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-03-28 16:09:03.126570: E tensorflow/stream_executor/cuda/cuda_driver.cc:936] failed to allocate 1.33G (1427963904 bytes) from device: CUDA_ERROR_OUT_OF_MEMORY
2018-03-28 16:09:03.126752: E tensorflow/stream_executor/cuda/cuda_driver.cc:936] failed to allocate 1.33G (1427963904 bytes) from device: CUDA_ERROR_OUT_OF_MEMORY

My question is how much GPU memory is required？？It's obvious that my computer is too weak.

[victordibia]

Glad it worked! Regarding memory, I have not done any GPU related tests. I ran the entire thing on my macbook CPU 16GB (no GPU). You can also try reducing the number of workers (to maybe 2?)

…

-V.

[gangyahaidao]

only use CPU, too slow? what's your fps?

[victordibia]

Please refer to the readme in the repository.

…

On Thu, Mar 29, 2018 at 08:19 gangyahaidao ***@***.***> wrote: only use CPU, too slow? what's your fps? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#14 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABea_-es_3sabG0xnKh1X_YOPYLaJpKNks5tjPuVgaJpZM4S7RgK> .

[TyrionZK]

after i reinstall tensor-flow v1.4.0-rc0, it works fine! thanks@victordibia
but i got such errors, when startup comes up CUDA_ERROR_OUT_OF_MEMORY

`rosrobot@rosrobot:~/git/handtracking$ python3 detect_multi_threaded.py
{'num_hands_detect': 2, 'score_thresh': 0.2, 'im_height': 180.0, 'im_width': 320.0} Namespace(display=1, fps=1, height=200, num_hands=2, num_workers=4, queue_size=5, video_source=0, width=300)
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
>> loading frozen model for worker
> ====== loading HAND frozen graph into memory
2018-03-28 16:09:03.123319: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-28 16:09:03.123576: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-28 16:09:03.123690: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-28 16:09:03.124384: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.55GiB
2018-03-28 16:09:03.124405: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-03-28 16:09:03.124661: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.55GiB
2018-03-28 16:09:03.124677: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-03-28 16:09:03.124862: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
name: GeForce MX150 major: 6 minor: 1 memoryClockRate(GHz): 1.5315
pciBusID: 0000:01:00.0
totalMemory: 1.95GiB freeMemory: 1.55GiB
2018-03-28 16:09:03.124896: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-03-28 16:09:03.126570: E tensorflow/stream_executor/cuda/cuda_driver.cc:936] failed to allocate 1.33G (1427963904 bytes) from device: CUDA_ERROR_OUT_OF_MEMORY
2018-03-28 16:09:03.126752: E tensorflow/stream_executor/cuda/cuda_driver.cc:936] failed to allocate 1.33G (1427963904 bytes) from device: CUDA_ERROR_OUT_OF_MEMORY

My question is how much GPU memory is required？？It's obvious that my computer is too weak.

How to install tensor-flow v1.4.0-rc0 ? I can not find this version with conda. Thanks.

[victordibia]

Hi @TyrionZK

One way to solve this error is to generate a frozen graph (from the model checkpoint I provide ) using your current version of tensorflow.

The tensorflow object detection repo has a python file for this.

You can copy it to the current directory and use it as follows

python3 export_inference_graph.py \
    --input_type image_tensor \
    --model-checkpoint/ssd_mobilenet_v1_pets.config \
    --model-checkpoint/model.ckpt-200002 \ 
    --output_directory hand_inference_graph

[TyrionZK]

Firstly, thanks for your reply. @victordibia

I have try it as you say. But some error happens as below. Then I try to solve the error with some guide about how to install tensorflow object detection API. After 1 day, it doesn't work. Can you give me some detailed instuction ? I will very appreciate it.

Traceback (most recent call last):
File "export_inference_graph.py", line 96, in
from object_detection import exporter
File "D:\ProgramData\Anaconda3\envs\detect\lib\site-packages\object_detection-0.1-py3.6.egg\object_detection\exporter.py", line 20, in
from tensorflow.contrib.quantize.python import graph_matcher
ImportError: cannot import name 'graph_matcher'

My environment:
win7
anaconda3
python 3.6.6
tensorflow 1.4.0