Train a Custom Object Detection Model using Tensorflow Lite Model Maker | Transfer Learning

Train and deploy a custom object detection model on Raspberry Pi. The video demonstrates preparation of your data including labelling of objects in the image, training the model and running the model on Raspberry Pi.
Read the article:-
helloworld.co.in/article/cust...
Using a Colab notebook (similar to Jupyter notebook), you can easily train a custom model for your use case. The original notebook by Khanh LeViet of Tensorflow team is available at this link:-
colab.research.google.com/git...
You can save a copy of the notebook in Google drive and personalize it for your use. The notebook I derived from the original notebook is here:-
colab.research.google.com/dri...
The code to run the custom model on Raspberry Pi :-
github.com/jiteshsaini/custom...
#TensorflowLite #modelmaker #customObjectDetection

[@Mike-dl6xe]

I was literally waiting for this video! Thank you so much!

[@japhethmumo2383]

Great tutorial! Keep it up. You have assisted me. Much thanks

[@jitesh_helloworld]

Thanks

[@EdjeElectronics]

Nice work on this video! 🙂

[@jitesh_helloworld]

Thanks a lot. I am your subscriber and watched all your videos. learnt a lot from them. This comment means a lot.😊

[@asmajassim7817]

Hi, I have already trained my model in yolov5s (on ultralytics) and exported it as a .tflite model. I tried to follow your code on deployment on raspberry pi but I keep receiving this error in the line: output_details=interpreter.get_output_details()[index] IndexError: list index out of range could you please help if you know what to do

[@coding_corner8210]

when I started installing tflite packages at start it took all the space on disk in meta data what should I do

[@andyhgnguyen]

Thanks so much for sharing. You mentioned in the video that the new result model from transfer learning can not recognize the pre-trained labelled objects. What if I would like to retain the capability to recogize existing objects while training extra objects with transfer learning? Can we do that?

[@kk-zv2vt]

Hi, I ran the exact same code and it works if I don’t use the coral accelerator, however when I run it with the coral I get tge following error: ValueError: Output tensor at index 0 is expected to have 3 dimensions, found 2. How can i solve this issue, please help

[@simpasalsunda7244]

Installing the !pip install -q tflite-model-maker-nightly !pip install -q tflite-support-nightly takes so long time, and disk space, how do i fix this?

[@melis4820]

thank you.

[@zivanaf]

Installing the !pip install -q tflite-model-maker-nightly !pip install -q tflite-support-nightly takse SO much space it actually make the Disk's space run out how can i solve this?

[@yo0seftal3at72]

i got the same problem

[@poojap7]

can you please make video on creating custom object detection model for tensorflow js

[@cojad]

Hi, As Colab now using python 3.10.x. Train_custom_model_tutorial.ipynb from Khanh LeVit is now fail to execute on current colab. Would you able to rewrite a working one for now? I'm having trouble even to start train model.

[@Hulkmb]

have you got any solution ...?

[@ptsd7403]

After running this code in the Colab: model = object_detector.create(train_data, model_spec=spec, batch_size=4, train_whole_model=True, epochs=20, validation_data=val_data) I got a error message: ValueError: The size of the train_data (0) couldn't be smaller than batch_size (1). To solve this problem, set the batch_size smaller or increase the size of the train_data.

[@onedori1]

have you found a solutuin for this?

[@mukat3222]

Hello! After running this code in the Colab: model = object_detector.create(train_data, model_spec=spec, batch_size=4, train_whole_model=True, epochs=20, validation_data=val_data) I got a error message: ValueError: The size of the train_data (0) couldn't be smaller than batch_size (1). To solve this problem, set the batch_size smaller or increase the size of the train_data. i have 64 trained Pictures with XML files and around 10 Validate Pictures with XML files, i did bounding boxes via LabelImg. I was trying to lower Batch size in the code and also i was trying to train more pictures, with no luck. Do you know any solution?

[@ptsd7403]

Anyluck ?

[@mikhilpatel5619]

did you find a solution? I am having the same issue

[@phoque6]

Your path to the train folder is wrong so it is no able to find any images that is why they detect as 0.. Anyways with the new updated colab pythn 3.10, tflite model maker is not able to install properly .Everyone seems to face this problem recently. i hope tensorflow resolves this soon

[@behindafact7535]

what camera did you use?

[@jitesh_helloworld]

Raspberry Pi cam

[@user-df9cq2ue2g]

Hi, thank you for this video. One question. How can I create a label.txt file, which contains the labels and their ID?

[@jitesh_helloworld]

It is created manually. You can use any editor for that. I created it using simple Notepad. Read more here helloworld.co.in/article/custom-object-detection-model

[@user-df9cq2ue2g]

@@jitesh_helloworld Great, thanks for your quick reply and help.

[@ptsd7403]

Hello sir please help there is an error with the code that ValueError: The size of the train_data (0) couldn't be smaller than batch_size (4). To solve this problem, set the batch_size smaller or increase the size of the train_data. what is the issue

[@shihaoshen2914]

hi, idk if u still need the solution. But here's my comment: In "Step 1: Load the dataset", make sure the train and validate folders being called are the ones u just uploaded. For example, I prepared my datasets with folder name 'val' instead of the default one they were using 'validate'. Then in here i should do 'custom_data/val' but not 'custom_data/validate', etc.

[@ajitgadkari127]

I was waiting for this...thanq so much ...can u please include this model in human following as well so tht we can detect one of the object and follow.

[@jitesh_helloworld]

I am thinking of that. 👍

[@techy9912]

Have you try on large datasets? Please help

[@jitesh_helloworld]

It should work. Rather larger dataset would help in training more accurate model

[@user-cz4iz3kn4c]

Hi I am running out of space in google collab while installing tflite-model-maker-nightly. Is there anything which I can d other than upgrading to collab pro.

[@jitesh_helloworld]

Thanks for pointing out. I also faced problem with its installation. May be some changes in the package by the Google team. Need to explore how to install tflite-model-maker. Rest of the script should remain the same. Let me know if you find a way ahead.

[@shihaoshen2914]

In CoLab: ctrl+shift+P. "use a fallback runtime version"

[@DSET_ChiragNRaj]

@@jitesh_helloworld Hey hi, First of all thank you for the vedio and even Iam facing the same problem like disk space is full in the allocated storage. Please let us know if there is some solution to over come this problem because i am using this for my project.

[@muhammadmurtaza4287]

@@shihaoshen2914 it is not available. what is solution?

[@phoque6]

@@DSET_ChiragNRaj this is recent problem with google colab updated python3.10. Even the use fallback runtime version" solution doesnt work anymore :( lets hope tensorflow resolves this fast

[@leleyu2682]

now the model maker colab notebook doesn't work, while trying to install, do you have any other solution for this?

[@Hulkmb]

have you got any solution ...?

[@egrabhishek]

Not working for me too

[@leleyu2682]

@@Hulkmb yes I got the solution

[@leleyu2682]

@@egrabhishek try creating virtual env in colab and then install the specific version. its working for me now

[@egrabhishek]

@@leleyu2682 great. Thanks. Will give it a try.. have a nice day ahead

[@ameer-alahmadi]

Thanks a lot for your great tutorial . But, can you please explain how to convert yolov8 to tflite model and use it with raspberry pi + coral usb accelerator?

[@jitesh_helloworld]

I have not tried that but google search shows few article for such conversion. One such article is www.google.com/url?sa=t&source=web&rct=j&url=github.com/NSTiwari/YOLOv3-to-TensorFlow-Lite-Conversion

[@japhethmumo2383]

You may also export the yolov8 dataset into pascal voc format and use the concept out teacher has used

[@ashikkirmani9070]

I want to make bird detection. is that possible

[@beyzaboz8446]

error: subprocess-exited-with-error × python setup.py egg_info did not run successfully. hello I give that error can you help me please? how can I load !pip install -q tflite-model-maker-nightly !pip install -q tflite-support-nightly

[@jonatapaulino]

Thanks for the video, it helped me a lot. I've already subscribed to the channel. Thanks. I have 1 question: I annotated all my images in roboflow, and I can export my database through a code that the roboflow website itself provides. In this case, step 3 would be different. How would I go about adapting? Thanks.

[@jitesh_helloworld]

Glad you found it useful. Im not sure how roboflow works. For this project, you should be able to annotate your images and create a corresponding xml file for them using any tool you like. Hope roboflow allows that

[@alialaa5726]

Hello, I also used roboflow to annotate my images what did you do in step 3 I would appreciate you help.

[@jonatapaulino]

@@alialaa5726 I haven't done the third step yet, because I wanted to import the code directly from roboflow like I did in other projects of mine.

[@alialaa5726]

@@jonatapaulino can you tell me about a tutorial that shows how to import directly from roboflow

[@protimaranipaul7107]

DOES NOT WORK.........

[@engammar1794]

There a major error when i run the pip installs : sub process errors. Any way to overcome this issue? !pip install -q tflite-model-maker & !pip install -q tflite-model-maker-nightly But both statement finally show a error while installing the dependencies. And here it is: error: subprocess-exited-with-error × Getting requirements to build wheel did not run successfully. │ exit code: 1 ╰─> See above for output. note: This error originates from a subprocess, and is likely not a problem with pip. Getting requirements to build wheel ... error error: subprocess-exited-with-error × Getting requirements to build wheel did not run successfully. │ exit code: 1 ╰─> See above for output. note: This error originates from a subprocess, and is likely not a problem with pip.

[@hung1853]

Me too. Did you solved this problem ?

[@duartesomsen6856]

@@hung1853 Have you found a solution yet? Ive been busy for the last 4 days trying to get this stuff to work.....

[@ilyesyousi6911]

@@duartesomsen6856 your are fix the erreur !?

[@muhammadshahab5749]

Sir ye google cural usb tu official Website se nahi melta kha say mel jaye ga And great Video late but always great contant 👍👍👍👍

[@muhammadshahab5749]

q k mujhe Google Coral usb chahenyee

[@jitesh_helloworld]

If you know someone in US / UK ask them to buy and ship it to you

[@jitesh_helloworld]

Thanks

[@akhilkp5378]

Installing the first first two packages !pip install -q tflite-model-maker-nightly !pip install -q tflite-support-nightly taking the entire memory how to solve it pls help

[@VlinGood]

Same problem here. Please share solution if anyone has one?!

[@valentinepolyakov3506]

@@VlinGood Same problem