The TensorFlow models download by download_iter02_tensorrt.sh are actually ONNX models
daviszhu88 opened this issue · comments
Issue Type
Bug
OS
Windows
OS architecture
x86_64
Programming Language
C++
Framework
TensorFlow
Model name and Weights/Checkpoints URL
https://github.com/PINTO0309/PINTO_model_zoo/tree/main/284_CREStereo
Description
The model files are actually ONNX files. Could you have it fixed as soon as possible? Thanks and I really appreciate it.
Relevant Log Output
No response
URL or source code for simple inference testing code
No response
onnx2tf -i xxxx.onnx -osd -cotof
Thanks for your quick response. I tried to use your tool to do the
conversion but had problems (see the below). Does the tool require Python
3.10? I have Python 3.11. I am retrying with the command you provided.
(CREStereo_ONNX) C:\Hongwei\DeepLearning\onnx2tf>onnx2tf -i
crestereo_init_iter20_720x1280.onnx
File
"C:\Users\davis\AppData\Roaming\Python\Python311\site-packages\tensorflow\python\framework\importer.py",
line 515, in _import_graph_def_internal
raise ValueError(str(e))
ValueError: Invalid GraphDef
Also, I noticed that the ONNX files have different names. Do you convert
the MegEgnine model to ONNX, or convert from the PyTorch model to ONNX?
How can I specify the iteration number and image size during the conversion?
Still, if you do have the TensorFlow models already converted, it might be
more convenient for me to just get the models from you, just like the ONNX
files.
Do you ever tried to use those models in C++?
Thanks and I really appreciate your time and kind help.
Hongwei
ESRI (www.esri.com), California, USA
…On Tue, Jul 25, 2023 at 4:35 PM Katsuya Hyodo ***@***.***> wrote:
onnx2tf -i xxxx.onnx -osd -cotof
—
Reply to this email directly, view it on GitHub
<#351 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AS6UABLUX3PIKAWZ6QOBKDTXSBJ4JANCNFSM6AAAAAA2XRGMP4>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
onnx2tf -i xxxx.onnx -osd -cotof
Took very long time (>1 hour) and get an error:
(CREStereo_ONNX) C:\Hongwei\DeepLearning\onnx2tf>onnx2tf -i crestereo_init_iter20_720x1280.onnx -osd -cotof
File "C:\Users\davis\AppData\Roaming\Python\Python311\site-packages\tensorflow\python\framework\importer.py", line 515, in _import_graph_def_internal
raise ValueError(str(e))
ValueError: Invalid GraphDef