search

nicknochnack / TFODCourse

961 stars 923 forks source link

tensorflow.python.framework.errors_impl.AlreadyExistsError: File system for s3 already registered #56

Open HH0000HH opened 2 years ago

HH0000HH commented 2 years ago

When running the command line python Tensorflow\models\research\object_detection\model_main_tf2.py --model_dir=Tensorflow\workspace\models\my_ssd_mobnet --pipeline_config_path=Tensorflow\workspace\models\my_ssd_mobnet\pipeline.config --num_train_steps=2000

I am getting the following error image

lucas-colares commented 2 years ago

Me too!!

digvijaydhanker commented 2 years ago

were you able to figure out a work around for this issue?

digvijaydhanker commented 2 years ago

So I figured this out dont install tensorflow 2.4.1 instead do 2.8.0 and when you do that you will get a bunch of warnings to install 10 or 12 libraries, pip install all that and then it'll work

yzn-lhsn commented 2 years ago

So I figured this out dont install tensorflow 2.4.1 instead do 2.8.0 and when you do that you will get a bunch of warnings to install 10 or 12 libraries, pip install all that and then it'll work

If you don't mind, could you share what versions you have for: Python, Cuda, and cuDNN. I originally went with 3.7.3, 11.0, and 8.0.4

davemaster commented 2 years ago

The examples works perfectly with Tensorflow 2.8. Keep in mind, in WHAT PATH are your notebook or where are you copy/extracted the TFODCourse files... check that very carefuly

yzn-lhsn commented 2 years ago

The examples works perfectly with Tensorflow 2.8. Keep in mind, in WHAT PATH are your notebook or where are you copy/extracted the TFODCourse files... check that very carefuly

Yep! The following fixed this issue for me: !pip install tensorflow==2.8.0 tensorflow-gpu==2.8.0 --upgrade

However, I also needed to upgrade to CUDA 11.2 and cuDNN 8.1 to follow requirements here.

EDIT: Never mind, Pytorch does not supprt CUDA 11.2 sigh... I'm trying to redo everything with CUDA 11.3 and cuDNN 8.2.1

What versions did you use?