Train a Deep Learning model for custom object detection using TensorFlow 2.x (On Google Colab)

This video shows step by step tutorial on how to train an object detection model for a custom dataset using TensorFlow 2.x. The custom object trained here is a face mask.

① ⚡⚡ My Website Blog post on this ⚡⚡

② ⚡⚡ My Medium post on ⚡⚡

③ ⚡⚡ My Colab notebook for this ⚡⚡

④ ⚡⚡My GitHub link for this ⚡⚡

⑤ ⚡⚡ My labeled dataset for this (PASCAL_VOC format XML files)⚡⚡

⑥ ⚡⚡ Article on Dataset Labeling/Annotation⚡⚡

#TensorFlow #customobjectdetection #mobilenet #TF2 #tensorflow2 #maskdetection #SSDModelTraining #facemaskdetection #objectdetection

▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬

▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Making these videos takes a lot of time and effort, so if you like these videos and if you can, then please support the channel using any of the following:

► Buy me a coffee! ☕

► Support channel on Patreon! 🖖


▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Thanks for watching!

[caraburrichter]

A big thank you for your tutorial, I am really glad and thankful I found it. I used it for a custom object detection for a university project and after a few adaptations, it worked great. Greetings from Germany

[elvenkim]

Very glad to finally found 1 tutorial that is idiot-proof. Very well-explained and I no need to scratch my head running at different Tensorflow pages!!

[atmavidyavirananda]

The most comprehensive tutorial I could find on SSD implementation with TF2. Really helpful, thank u so much!!

[surajojha]

wow! i just trained my first object detection model. THank you

[shivamrana]

Thanks sir for doing this muchh efforts for us.... u were working from 4 am to 7am for this video....Really appreciative sir.

[franciscovelasquezferrada]

Hello my friend, thank you for sharing your colab, it's really well explained :D

[HerohollandYT]

Thank you very much for such a useful and easy to follow tutorial on this topic!

[hmt.sun]

Sir I am a PhD scholar and ur video is going to help me for buliding a ml model for intelligent Vehicle system.

[marcomolino]

thank you very much man, really a great job!!

[lmrzmvx]

when try to trained coustom model it gives error in step 15 as no attribute name case in control_flow_ops.case line no.453

[dribletdrop]

i have take this error at step 14 : "No module named 'tensorboard ", help please

[qeuqqtd]

Thank you so much for such an amazing and detailed video, I got an error when I started the training module has no attribute 'case' could you please help me with this, that would really a great help.

[sarahm]

Great tutorial!! Fantastic! How can we view and plot accuracy?

[conor_codes]

Great video so far, though I am running into issues. First, I am having a Library DNN issue for step 15. I have pip installed tensorflow 2.10 and restarted runtime, though no luck. How do I fix this problem?

[Kashut-flcu]

Hi, thank you for the tutorial!
Unfortunately, I have encountered a problem when training the model (at step 15a) :-

Node:
OOM when allocating tensor with shape[15052800, 1024] and type float on by allocator GPU_0_bfc
[[{{node
Hint: If you want to see a list of allocated tensors when OOM happens, add to RunOptions for current allocation info. This isn't available when running in Eager mode.

May I know how I can solve this?

[hriticklohani]

Thank you very much, sir. I did all steps but at last on testing the image I m just getting the image without any bounding box and labels in masked or unmasked people.

[RRAMELL]

Thank you very much for the tutorial, sir. It really helped me. But I need to deploy the model to a website. Can the resulting model be applied to the website? Because other video tutorials on youtube not really help me. Hope you can respone..

[hasancansolakoglu]

my training images was not the same size so now my exported model doesnt have bounding boxes. how can i fix?

[GoodMuffinReal]

hey, when I try to run the final code to see if it can detect the classes in some images, the code gives me this error:
ValueError: Python inputs incompatible with input_signature:
inputs: (
tf.Tensor(
[[[[ 0 20 45 255]
....
[ 0 4 8 255]]]], shape=(1, 1080, 1920, 4), dtype=uint8))
input_signature: (
TensorSpec(shape=(1, None, None, 3), dtype=tf.uint8, name='input_tensor')).

Do you think you can help me?

[dany]

I tried running this notebook and got two types of errors:
1. The module does not have the 'case' attribute.
2. The training interrupts with '^C'.

I solved both in this way:
1. by running command !pip install tensorflow[and-cuda]==2.13.0
2. decreasing the batch size from 64 to 32

thanks to @techzizou for this tutorial i appraciated a lot