Tensorflow 2 Custom Object Detection Model (Google Colab and Local PC)

Train your own custom object detection model with Tensorflow 2! Choose any object you like and follow along with this tutorial! After watching this, you'll be able to train your own model either in the cloud using Google Colab or on your PC.

Other videos in this series:

*Chapters:*
00:00 - Intro
*Prep Work*
00:51 - Prepare images
07:40 - XML to CSV
09:06 - Generate TF records
12:38 - Download a base model
14:50 - Edit model config file
*Local PC Training*
21:57 - Start training
26:35 - Tensorboard (kind of)
*Google Colab Training*
29:31 - Intro to Colab
30:55 - Run TF code blocks
32:02 - Use Kaggle to get our data
36:25 - TF records and config file
41:13 - Another more different tensorboard
41:45 - Start the training
* Using the New Model*
44:35 - Export inference graph
49:50 - Resume training from checkpoint
51:56 - Use new model
56:26 - Outro

[rlho]

THANK YOU! this helped me so much, even chatgpt got crumbled up because he doesn't know the newer versions. you're a life saver my guy

[muzakkirali]

Followed many tutorials but your videos were simple and organized, your videos helped me complete my college project. Thanks a lot!!!

[rajivpannuri]

This is the best and clean tutorial I have seen so far. Thank you.

[BeLKa]

THANK YOU OH MY GOSH, YOU RE MY HERO

just thank you for all this playlist - you`re making a world a better place by helping a newcomers like me to deal with so much frustration

I`ve spend HOURS trying to implement things from tutorials but only after watching your videos (at least 3 times each one XD) I`ve finally understood what going on here with all these models, apis, label and etc.

geez, really I cant thank you enough (should probably buy you a coffee one day, when I`ll have more stable financial situation as a gesture to show my appreciation)

[hexteriaenrage]

Awesome! Thank you so much for making this!

[HumptyDumptyActual]

This was a great video. Thank you for uploading.

[AshishBangwal]

you won't belive how much you helped me, there are some simpler object detector trainers like Yolo's ultralytics but i was bound to use TF framework and your video helped me a lot, thanks and keep the good work up. ❤🙏

[zulnaim]

hi can i know how you evaluate the trained model you create, like what code to use to evaluate using our val data

[pepega]

Hi ben, how do you get the traindata set on the local pc ? For example, after i train the custom dataset on my local pc . How do i get the train file ?

[VNNY]

Hi, I was wondering if I can use roboflow datasets instead of kaggle? if yes, how?

[dwikilazzaro]

Sir how can i get the accuracy to be tested?, or there any method to get the accuracy on the tensorboard?

[EED-qjtv]

I want to use faster rcnn with mobilenet. So can i change the backbone or something else using variables and checkpoints?

[neilpatrickdiscaya]

Thank you for this tutorial. However, during training I encountered the following error:
TypeError: Since we are restoring a v2 style checkpoint restore_map was expected to return a (str -> Model) mapping, but we received a (str -> CheckpointV1) mapping instead.

I am training it it Local

[NarayanaC-mpwc]

Hi, I am stuck at a point where while I was training a custom model my execution stopped or paused at

2024-06-19 12:33:22.948786: W Input of GeneratorDatasetOp::Dataset will not be optimized because the dataset does not implement the AsGraphDefInternal() method needed to apply optimizations.

Can anyone help me on how to resolve this. I'm using tensorflow==2.13.1

[ameer-alahmadi]

Can you please explain how to convert the pb model to onnx and use it with opencv. Thanks a lot.

[AmochaSagolsem]

I got this error "ValueError: Cannot assign value to variable ' Shape mismatch.The variable shape (1, 1, 1024, 512), and the assigned value shape (1, 1, 1088, 256) are incompatible." when trying to train using faster RCNN. in the train and test image. the image size is 295x440. is it because my images are too small?

[sanjoetv]

you are showing about the config files but we dont have access to those.. where can i get those config files

[sanjoetv]

please answer where can i access the config fiels that you edit

[nlwinacties]

In which file i can perform some action while a model is detected? Let say i want to send a email to someone if a person is detected?

[katana]

instead of using kaggle to save, can i use drive directly to save all the steps