Wrong catenation type leads to model exportation failure [TF2 + object detection]

[young24]

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• I am reporting the issue to the correct repository. (Model Garden official or research directory)
• I checked to make sure that this issue has not already been filed.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md

2. Describe the bug

I was following the official TF2 object detection tutorial. However, I failed to export the trained model. The issue seems to be that tensorflow used '/' instead of '' when creating some temporary file.

3. Steps to reproduce

the official TF2 object detection tutorial. I was using the pre-trained model "faster_rcnn_resnet101_640x640".

4. Expected behavior

successfully export the trained model

5. Additional context

6. System information

• Windows 10 Enterprise
• TensorFlow installed from (source or binary): Anaconda 3
• TensorFlow version (use command below): 2.3.0
• Python version: 3.8.3
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version: 10.1
• GPU model and memory: GeForce GTX 1660, 6 GB GDDR5

[young24]

I managed to export the model "EfficientDet D1". So I guess it's a model-specific issue.

[ravikyram]

@young24

Glad to know that it worked for "EfficientDet D1"
Will it be possible for you to share colab link for "faster_rcnn_resnet101_640x640". model so we will try to reproduce the issue in our environment.It helps us in localizing the issue faster.Thanks!

[young24]

@young24

Glad to know that it worked for "EfficientDet D1"
Will it be possible for you to share colab link for "faster_rcnn_resnet101_640x640". model so we will try to reproduce the issue in our environment.It helps us in localizing the issue faster.Thanks!

Sure. I'll try to implement it in Colab. Thanks for your message.

[young24]

Hi @ravikyram , I implemented the training pipeline in colab (check it out here). However, this time worked fine. I reduced the num_steps from 30000 to 100. Is it possible that this issue is due to the local Windows machine I'm using?

Another difference is that in colab, I used a different command to export the graph. I dropped "input_type". I'm trying to reimplement the training pipeline on my local machine. I'll inform you once it's done.

[young24]

I tried to reimplement the pipeline on my local machine and used the same command to export the model as in the colab. However, I got the same error message. Since it's this "", "/" issue, I guess it's Windows-specific.

[ravikyram]

@young24

Request you to grant me the access for colab link.
It looks the issue is more specific to your system.Thanks!

[young24]

@young24

Request you to grant me the access for colab link.
It looks the issue is more specific to your system.Thanks!

Hi Ravikyram, thank you for your time. I have granted you the access by your email address.
Just let me know if you need anything from me.