Install Tensorflow Object Detection From Scratch in 5 Steps | Python Deep Learning

Running into errors?

Tired of seeing "ERROR: XXXXX something didn't go right"

I hear you!

Installing Tensorflow Object Detection can be a little tricky at times. But it doesn't need to be. In this video we'll run through the 5 key steps you need to go through in order to setup the Tensorflow Object Detection API. We'll go from a bare install without Python to a fully setup machine.
0:00 - Start

In this video you'll learn how to:
02:21 - Install Python using Anaconda
05:51 - Install Microsoft Visual Studio and C++ Build Tools
09:01 - Install CUDA and CUDNN for NVIDIA
15:04 - Install Protoc for Protocol Buffers from the official Github Repo
18:58 - Install Tensorflow Object Detection API for use with Python

Links Used:

Oh, and don't forget to connect with me!

Happy coding!
Nick

P.s. Let me know how you go and drop a comment if you need a hand!

[sharveensmith]

You're a lifesaver my dude, exactly what I was looking for. Love the way you make it easy to understand and go through step by step

[matt-cqnd]

Thanks for this clear and straightforward tutorial! Great work

[edjai]

Hi @Nicholas, have another question:
You mentioned in your tutorial a YouTube folder by the minute 24:16, within you install Jupyter and run a tutorial.
I am not sure if I have understood this part correctly.
Is the YouTube folder and the Jupyter tutorial installed altogether with the python -m pip install . command??
Thanks again

[dan]

The instructions were super easy to follow. This video saved me days of research time of how to get this up and running.
I was able to make it work in 1.5 hrs.
Great content as always! Thank you!

[shubham-ppcw]

Even though video is half an long but installing all the others file will take more than double the time of the video. However is simple enough such any one can perform or install . Thank you for making such good video

[elinamalthea]

Hi! I am currently using windows 11, what CUDA and CUDNN version do you recommend to download? Thank you

[chadaravalliuttej]

At 20:10 should we navigate to D drive or we can also paste the clone model with out navigating to D drive

[SolidClouds_NL]

ERROR: Directory '.' is not installable. Neither 'setup.py' nor 'pyproject.toml'

HELP :S

[markbenedictdjarabese]

Greetings sir! Thank you very much my training is now working using GPU. I am having a hard time searching. I switch from windows to linux then go back to windows. I have encountered issues with the cuda then I have read your comment downgrading from tf 2.4 to 2.3.1 solves my problem. Right now I am doing real time sign language recognition system that involves basic words and phrases in FSL. Again thank you very much for making this video . Please stay safe and God bless your family sir

[yokeshyokesh]

hi nicholas, I'm facing a problem at the final step "python -m pip install ." throwing an error "ERROR: Could not install packages due to an OSError: [WinError 206] The filename or extension is too long", Can you help me resolve it :)

[muralimanohar]

sir, while I was downloading the tensor flow API some of the commands are displaying errors in the command prompt...how can I solve that???

[kartikk]

Hey @Nicholas,
I've tried multiple times and it always says "Failed building wheel for pycocotools" when I start the installation. Any idea what the issue might be?

[turkishfail]

you are a legend man. Keep up the great work.

[gopikrishna]

Hi Nicholas,
I am doing this same topic for my project
But the problem is I don't have nvidia gpu in my system, it has Intel r hd graphics 5xx, will this still work on my system.
I heard installing TensorFlow-directml package will solve the problem but will their be any changes that has to be made in the code.

[zainabalbakaa]

Hi Nicholas, thank you for this video, , I have error in python -m pip install the error is ERROR: Could not build wheels for opencv-python, opencv-python-headless which use PEP 517 and cannot be installed directly what i have to do, , can you help me

[IgorMeloS]

Hey Nicholas, thanks so much for this video.

Nicholas, I have a problem during the installation. I'll try to explain you. Before to install the TensorFlow Object Detection, I've installed the TensorFlow 2.1 by source, due to the Compute Capability of my graphic card. When I've installed the TensorFlow Object Detection, my TF version was changed into 2.6, this version requires other CUDA configurations. I've reinstalled the TF 2.1 to run with CUDA, but the Object Detection API does not work. Do you have any suggestion to this issue ??? Because I tried many times, but without success. Thanks you for all your videos, this help-me a lot.

[alvarohenriquez]

This is a great video. For those who have experienced the 'cudart64_110.dll' error, its because you are installing Tensorflow 2.4.1 so the cuda and cudnn versions must be updated. This is the easiest way I think:

1. Activate your conda object-detection environment - 'conda activate name_of_env'
2. Issue 'conda list' command and check Tensoflow, cudatoolkit, and cudnn versions. If its tensorflow 2.4.1 the cuda stuff is probably not listed.
3. For Tensorflow 2.4.1 you will need cudatoolkit 11.0.3 and cudnn 8.0.5.39
4. If cudatoolkit and cudnn are installed, but with lower versions, you will need to remove them
a. conda remove cudatoolkit
b. conda remove cudnn
4. Install from the nvidia conda channel
a. conda install --channel nvidia cudatoolkit=11.0.3
b. conda install --channel nvidia cudnn=8.0.5.39
5. Close your terminal session, then open a new one, go to your environment and test your install.
python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
6. The first line of output should be:
2021-02-16 19:55:16.628819: I Successfully opened dynamic library cudart64_110.dll

With this, I am able to run different versions of cuda in different environments.

[tahiraleem]

Hi Nicholas, I'm installing tfod in Nvidia Jetson Nano which uses Debian os. Therefore I'm using VSCode. I want to know how can I install Visual C++ build tools using VSCode.

[icyvolts]

For some reason the ls command is not working for me, I have tried adding git bash bin to path and its other folder but it still is not working niether is the cp command

[eldridgerexroy]

hi can you help me, im stuck in step 6 when i need to train the models it give me this error
W Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found

i already install the cuda and cudnn,
but weirdly on step 2 my tf record both train and test is empty

Successfully created the TFRecord file:
Successfully created the TFRecord file: