0

I have been training an object detection model using ssd_mobilenet_v2_coco_2018_03_29.

Since last week, my colab .ipynb training script would give out mAP and AR of zero all the way till the end of training. [Picture of mAP and AR outputting zeros during training][1]

Until recently, I have successfully trained different versions of models without problems using the attached jupyter Training_Own_Data_Old.ipynb file below.

I checked the revision history to see if any changes were accidentally made, but there aren't any drastic changes except those mentioned below.

Some noticeable differences

Last week, I experienced the message- "Colab is experiencing issues connecting to Drive, and we are actively Investigating" whenever I try to connect to colab.

I also noticed that python-pil got updated. I installed the old version and tried again but that didn't solve the problem.

I've also checked the pipeline config files, training logs, tried with/without using dropout, just running the old .ipynb file itself without changes but still no luck.

I've attached github of the old working Training_Own_Data_Old.ipynb file an the new Training_Own_Data.ipynb file with errors. Logs are within each .ipynb file.

https://github.com/WalterYeYint/Jupyter-files

Could somebody please help? Any help is appreciated. [1]: https://i.stack.imgur.com/7oNT0.png

Electrinos
  • 1
  • 3
  • Welcome Electrinos! Can you try to be more focused in your question? Try to pinpoint the issue yourself first to make it easier for other members of the community to help you – 465b Aug 01 '20 at 12:40

1 Answers1

0

[SOLVED]: My script has been pulling tensorflow models repository everytime it runs. Downloaded that repository from the last time my script was working solved the problem.

Electrinos
  • 1
  • 3