On-device object detection: Train and deploy a custom TensorFlow Lite model

preview_player
Показать описание
This video is part of a learning pathway that teaches you how to do object detection on mobile.

In this video, you’ll learn how to train a custom object detection model using TensorFlow Lite Model Maker and deploy it to an Android app using TensorFlow Lite Task Library.

Рекомендации по теме
Комментарии
Автор

How you gather/label the data is maybe the least covered - important step in machine learning. 95% of learning material assumes you won't use it in the real world and therefore never teach you how to use real-world data. That is why I like this video a lot - it includes this part(not in full but enough for me).

socrates_the_great
Автор

I am labeling images with LabelImg and with every picture i have, i get 1 xml file. How do i put all those XML files in just one CSV to load it into my model.
Here you use csv file from gs://cloud, how can I make a local csv file with the images I have?

davidhe
Автор

Can we have evaluation function for object detection models in tensorflow zoo 2 also. Like for every class, the eval file only show mAP for all, not for every object individually.

zoroXgamings
Автор

I am trying to show thew results in listview instead of on image with boundingbox for testing purpose. how do i do that?

developerandroid
Автор

that optional testing part is what i needed

Raj-szpg
Автор

I'm getting this error "required broadcastable shapes [Op:Mul]". on "model.evaluate_tflite"

gracenotesheaven
Автор

I've changed the number of detections in the efficientnet model (tflite_max_detections) before I train and also in Android Studio with (.setMaxResults) in the ObjectBuilder method but still the maximum number of detections is stuck at 25.

Anyone any solutions?

GlentoranMark
Автор

Custom object detection is working but when I click the camera button .... nothing happens. Any clues ... I am using a Pixel 4a and would like to use the camera

wintoncape
Автор

can u add a pick image from gallery module for this project?

kenshinx
Автор

None of the links in the description working

abdullahmohammed
Автор

Using colab, the first "required packages" takes so much time and space it actually fills up the Disk lol
whats up with that

zivanaf
Автор

Helll anyone can u plz answer this question what samples are used to decide when to stop training to avoid overfitting

sravyaannapureddy
Автор

Do you have any tutorial in recognizing images with accuracy and voice output?

joycemalubay
Автор

This might seem like a weird question but at what point when using Model Maker do you preference the headless model you're using? is it when you select your Model Architecture (EfficentDet-Lite2 etc.)?

WIIIIILLLLL
Автор

None of the links in the description working..

tumervural
Автор

Hello sir, i want to ask. How we can get the validation accuracy during training?

wahyuadiprastya
Автор

How to add multiple tflite model on one app

joshuamawutorklint
Автор

Good tutorial sir! I salute you! Do you have any guide on how you load dataset from local computer?

japhethmumo
Автор

Followed this tutorial for a custom dataset where images were stored in a local path. Doesn't work. I'm getting an error while importing the CSV of the dataset that says, "Image format not JPEG", despite the images being in the said format. Have tried doing it for both, .jpg and .jpeg extensions but still, the error persists.

nitintiwari
Автор

I labeled my dataset on roboflow, I've saved it on my drive. then I tried to load it from my drive
when I get into step 3 to train the model I get an attribute error that goes like: 'AttributeError: 'NoneType' object has no attribute 'label_map'..
please can somebody help me?

jackyguilhomme