Train a custom object detection model using your data

Рет қаралды 192,958

Күн бұрын

Learn how to train a custom object detection model for Raspberry Pi to detect less common objects like versions of a logo using your own collection of data.
00:00 Introduction
00:49 The 3 steps of training a custom model
01:24 Step 1: Create a training dataset
04:01 Step 2: Train a custom model with TensorFlow Lite Model Maker
09:03 Step 3: Deploy the custom model to Raspberry Pi
11:08 What’s next
Colab notebook to train a custom object detection model → goo.gle/3ocbqmI
Sample app to run the object detection model on Raspberry Pi → goo.gle/3GaABw3
Android figurine dataset → goo.gle/31DtXPL
Explaining “average precision” → goo.gle/3lBR5p6
Transfer learning → goo.gle/3pBDiAh
Responsible AI → goo.gle/2QEEuVV
Choose an object detection model architecture for Raspberry Pi → goo.gle/3lDe9DO
Make object detection run faster by using Coral → goo.gle/3EuH3xn
Watch all Machine Learning for Raspberry Pi videos → goo.gle/ML-raspberrypi
Subscribe to TensorFlow → goo.gle/TensorFlow
#TensorFlow #MachineLearning #ML #RaspberryPi #EdgeAI
product: TensorFlow - TensorFlow Lite, TensorFlow - General; fullname: Khanh LeViet;

Пікірлер: 213

@silvanovich Жыл бұрын

Thank you for sharing this straight forward approach to object detection. I have read book chapters and looked at a variety of web pages describing how this can be done, and while I am grateful to those who shared their knowledge, this video was at a level that I can understand. It explains the concepts without resorting to jargon and having to use a .tfrec input format.

@josenoventaeum3219 3 ай бұрын

You're so objective and clear that everything sounds veeeery easy!

@denizcanturk8441 Жыл бұрын

Great explanation and most importantly whatever he said works properly... Great job and thanks..

@Auditor1337 Жыл бұрын

thank you a lot, you made understanding this a lot easier.

@hj2931 2 жыл бұрын

thanks for sharing knowledge!

@richarballesteros5442 2 жыл бұрын

i have been 2 years trying to do this and finally works

@Lolatyou332 2 жыл бұрын

I've been using OpenCV for a professional usecase on object detection. I might try using tensorflow now that there seems to be more information available on implementation, such as this video.

@ChrisHalden007 2 жыл бұрын

Really cool. Thanks

@TechCree 2 жыл бұрын

I am really enthusiastic about your contribution and hope for some more videos of this kind. My goal is that the Pi can later output an acoustic message when it has detected an object. For example, if a person were to enter the room in which a blind person is located, I want the person to be recognized and I can use Python and possibly espeak to communicate that "Pete" has entered the room. Or at a danger point it would be possible to get a message when an object is approaching. Or, if necessary, switch lights on and off via GPIO interfaces or start an LED warning sequence. My head bursts when I think about the possibilities it offers. Please please continue in this direction !!!

@TechCree 2 жыл бұрын

I can find all recorded objects via the string dedetions (from your python script) and output them via the print command. I can now save it as a python command in a file or via api in a google sheet. the content is a bit caotic but I'll see how it gets along. please let me know if there is a better way. dn

@khanhleviet5416 2 жыл бұрын

@@TechCree To make your device speak out the detected objects, you can check out a Python library called gTTS from Google to convert text into speech. gtts.readthedocs.io/en/latest/

@shennaldsonarzaga4568 Жыл бұрын

Hello I'm dealing with the same project. Have successfully figured out the codes?

@alami9024 2 жыл бұрын

Thank you so much for this :)

@vladimirputin623 Жыл бұрын

Looks like this project stopped working, just getting a bunch of errors during installation

@jitesh_helloworld Жыл бұрын

Suddenly, model-maker package is not getting installed in the colab notebook you shared. Note book was working perfectly fine till last week.

@alzalame 2 жыл бұрын

Nice and easy , thank you very much

@user-cv4jm2sz5k 2 жыл бұрын

Does it work well without errors?

@avinashsoni9533 Жыл бұрын

how to keep .xml extension in dataset

@KarthickRobo Жыл бұрын

thanks a lot i am new to computer vison its easy to start

@user-ybk Жыл бұрын

공부하고 싶은 동영상이 여기 었었네요.. 잘 봤습니다~

@user-bj3vo4ct6l Жыл бұрын

Hi, great video. Is there a way to make the resolution of the pi camera video capturing clearer?

@nabeelmalik.4714 Жыл бұрын

When i execute !pip install -q tflite-model-maker-nightly It start downloading dependencies and never stop until disk low message has occured. Any solution to this problem.?

@ambient-videos Жыл бұрын

Same problem

@hichamm86 Жыл бұрын

Thank you so much for this video, please i have a question : how many classes can we train with tflite?

@margramiof7926 Жыл бұрын

This video series is pure gold. Thank you for sharing and explaining the power of this framework. Just a comment, the colab notebook in the above link has an error in the line which installs the required packages. Instead of : !pip install -q tflite-model-maker-nightly !pip install -q tflite-support-nightly as shown in the video. It says: !pip install -q tflite-model-maker !pip install -q tflite-support In my case, those commands took a long time (almost timeout ) and they did not work.

@jitesh_helloworld Жыл бұрын

Thanks a lot. It worked and saved me a lot of effort 👍

@luizabbadia3411 Жыл бұрын

Thank you so much!!! I built some frame work based on all that procedure and I was trying to make it work in the last couple of days and I was frustrated.Thanks a lot for posting this salvior post!!!

@Jinx-dt6sr Жыл бұрын

Been searching for a solution for like 3 hours now, Thanks so much!!

@elvenkim Жыл бұрын

thanks very much, i was stuck for days

@nabeelmalik.4714 Жыл бұрын

I execute both commands but it took very long time. Any solution for this problem..?

@bandara-sk3rf 9 ай бұрын

while I am downloading "!pip install tflite-model-maker" my Disk space is not enough for it. how can i manage it ?

@ungifted4311 3 ай бұрын

As of 31,Mar,2024, tflite model maker can not be installed, due to compatibility issue with Python version 3.10 (According several report in Github and other sites). Is there any solution ???

@muhammadnorirfan4957 2 ай бұрын

tell me if you find the solution

@user-ep3or2eg1j 2 ай бұрын

what is the problem to use older python?

@Terrar1988 Жыл бұрын

can you please make a video of creating an absolutely empty model and training it from scratch (not transfer), even if thousands of pictures are required? It would be very useful for our scientific studies, thanks!

@abhijitsutar3561 Жыл бұрын

I'm using the notebook currently in colab with python 3.9. While installing tflite-model-maker it's installing all the versions of tf-gpu >=2.6 , which leads to out of storage situation in colab, without properly installing the dependencies. If you checked my comment then please reply back . Thank you

@andreserafim1256 9 ай бұрын

Did you solve this? I'm having the same issue.

@shannirmala 8 ай бұрын

i'm having the same issue and no solution yet

@SanjeevKumar-dr6qj Жыл бұрын

thank you

@user-rj5tf9fq3e Жыл бұрын

Can't use your project for training know . Could you please fix the error of installation ?

@ltesGo 2 жыл бұрын

thx for your tutorial ! Can you make a video about training 3D object detection model （Tensorflow Objectron ）？

@khanhleviet5416 2 жыл бұрын

Thanks for the request. I'll look into that :)

@mugtabataha3355 2 жыл бұрын

Everything worked perfectly except for one thing when i downloaded my model and went inside examples/lite/examples.... I can't use cp - /download it says missing destination can you help me please

@abdullahimtiaz6564 2 жыл бұрын

Thank you so much for this. I have trained the model and it worked, now I want to take it to the next step like for taking some decisions. if a desired thing is detected then I want to have some controls with the pi, for example removing/sorting a product moving on a conveyer belt after it gets detected. How can we do it ?

@killua349 2 жыл бұрын

I want to do the same. Have you tried it?

@khanhleviet5416 2 жыл бұрын

It'll depend on what's the decision you want to take. For example if you want to remove a product moving on a conveyor belt, you'll need a robot arm or a mechanical device to do so. The robot arm or the device needs to have an API that allow you to control it from the Raspberry Pi.

@killua349 2 жыл бұрын

@@khanhleviet5416 yes I actually have a actuator , and i used the index which stored the value of the object detected in the code to further instruct. Thanks 👍

@eduardodeleon1132 Жыл бұрын

@@killua349 Could you explain to me, how did you manage to do it? Please. I am currently developing a task and I need to activate a solenoid valve when detecting a certain object.

@carlojhonsario1220 2 жыл бұрын

When I use print(detection_result.detections) I get these: [bounding_box { origin_x: 107 origin_y: 57 width: 439 height: 311 } classes { index: 0 score: 0.9921875 class_name: "surgical_mask" } ] How can I print the "class_name" only?

@Team-fh2rr Жыл бұрын

Good day! We have some issues with object detection on custom datasets and would be appreciated any advice. Is it possible to have a 1-hours consultation? How much will it cost?

@GeorgeLenoHolmesJr Жыл бұрын

Thank you for the video! Question: The way in which this model was trained assumes that there is always an 'android' in the FOV of the camera, because all of the provide images for training include an android in the photo (i.e. no null annotations are provided). Does Model Maker support null annotations? Is there a preferred Pascal VOC xml format that should be provided to account for null annotations?

@vikramm4967 Жыл бұрын

Hey, did you find any method to solve this?

@GeorgeLenoHolmesJr Жыл бұрын

@@vikramm4967 From my research it appears that tf lite is doing work in the background to train with null data. My team has trained models with null data by providing xml files that completely exclude the "object tag" (and its sub tags) from the xml file. (To be clear, we did not need to create a null data class or label). The trained model does well when deployed in identifying frames containing null data and not generating a bounding box. NOTE: It does appear important to test with the percentage of your training and validation data that includes null samples, for our application, a 50-50 split did not give good results; a 85-15 split, however, did provide good performance. Please let us know how this works for your application. Thanks

@vikramm4967 Жыл бұрын

@@GeorgeLenoHolmesJr Sure

@vikramm4967 Жыл бұрын

@@GeorgeLenoHolmesJr just to clarify, you created xml file without object tag and you did not leave background images without creating xml file Because i didn't create xml files for the background images and the model is still training, I just want to know if it is including these images or if i should create xml files

@GeorgeLenoHolmesJr Жыл бұрын

@@vikramm4967 Every image we trained with had a corresponding XML file. The images that did not have objects that needed to be detected, we used corresponding XML files without object tags.

@mardeenosman8979 2 жыл бұрын

Hello, thank you so much very good informations.. I have one question please Can we use pretrained model on different type of data for example if the data used was 2D images can I use transfer learning on 3D images??

@khanhleviet5416 2 жыл бұрын

No, Model Maker currently only supports transfer learning for object detection on 2D images.

@wasgeht2409 2 жыл бұрын

great

@lucapoli3562 Жыл бұрын

With this model can I use the command model.save to obtain the keras_metadata file? And I need to install a different version of Tensorflow to do this?

@shooby117 9 ай бұрын

the packages won't install on google colab

@aundo1647 9 ай бұрын

The colab notebook doesn't work anymore, !pip install tf-model-maker is stuck in an infinite download loop. Why is that?

@matt-cpsc Ай бұрын

7mo later, Apparently it's because of the version of Python that colab uses, you have to create a virtual environment and install an older version to be able to use tf-model-maker.

@defyphysics2955 25 күн бұрын

@@matt-cpsc how can i do that? how can i create a virtual environment on google colab and install an older version of Python? which version of Python do i need specifically?

@matt-cpsc 25 күн бұрын

@@defyphysics2955 search downgrade Python version, it's a guide from GeeksForGeeks

@Nicofisi 2 жыл бұрын

The code does not work anymore, even the linked collab doesn't work when you run it in the browser :(

@andrejustinvalenzuela2976 Ай бұрын

Can you do the model training on a windows pc instead of the RPi, in my case it is RPi 3b, then just download the finished model/product on the RPi after?

@zillur-av 2 жыл бұрын

I do not know what happened. When I run efficientdet0, it is working fine. But when I run a custom trained model, nothing shows and also no error.

@nachoridesbikes Жыл бұрын

hey ther i xould not run the example as tf.keras.mixed_precision.experimental has been deprecated. any help would be much appriciated! thanks!

@ramacharanchalla7996 4 ай бұрын

samples are used to measure how well the model can generalize and recognize objects that it has never seen before. Can anybody say

@user-xs3cm7oc6u Жыл бұрын

How to run custom binary image classification model converted to tflite in raspberrypi?

@python1depl183 Жыл бұрын

is it work for placard detection ...one placard contains 4 unique shapes while rotate..

@cristiandavidandrade807 2 жыл бұрын

Importe los paquetes necesarios. da error solucion ?

@thanyawach 2 жыл бұрын

I have a problem about training section and I cannot fix it

@peppiler 2 жыл бұрын

How can TensorFlow Lite models be containerized (using Docker)?

@jerecbazar205 10 ай бұрын

where Should I insert If I want to add audio in every object that is being detected

@russhall1097 2 жыл бұрын

Tried to use this with my own data. The format Pascal VOC, that must be different from my format, for Colab says there are zero images. In what format should we have our images? Jpegs and corresponding xml files of annotations? Is that Pascal VOC, or some other format?

@ayeshasiddiqui9068 Жыл бұрын

U should not use space in name of classes

@amayuruupanith1336 11 ай бұрын

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. I got this error.Can anyone solve this issue?

@babozone2 10 ай бұрын

저도 같은 문제를 가지고 있습니다. 구글은 텐서플로우를 관리할 생각이 없는거 같네요.

@thomashu1095 2 ай бұрын

I have moved to mediapipe model maker and successfully created a simple model to run on my Pi. However, I tried to compile it to a Edgetpu using your colab. After downgrade tensorflow from 2.15.0 to 2.13.0. It did compile, however, it failed to run on Pi because of some runtime errors. I noticed that there is a line “from tflite_support import metadata”. I wonder if there is an equivalent “mediapipe_support”? Would you please provide updated instructions? Thanks!

@mbrraghu Ай бұрын

hi, !pip install -q tflite-model-maker-nightly !pip install -q tflite-support-nightly is killing the colab space, more than 50 GB. What is the solution?

@yusufsezgi9958 Ай бұрын

Hello, were you able to solve it?

@jyothinkjayan6508 2 жыл бұрын

How many images of a particular class required for training?

@paarthvenomgaming1617 2 жыл бұрын

i am sorry but when i run this on my raspi-4 and my macos it shows an error where it says PortAudio not found.pls tell how to fix this error

@leonard866 2 жыл бұрын

Do you know if its possible to start TFLite on a MacM1?

@vishwasrb2961 4 ай бұрын

does it taking nearly 100GB to install the model maker? i mean to run the first step

@shaheermubarak 4 ай бұрын

yes did you found any solutions?

@vishwasrb2961 4 ай бұрын

@@shaheermubarak no

@davehitesh9697 Ай бұрын

hello today i saw this video i dont understand one things what is validation data ? i have to create a model for full face detection only face without any background

@elmehdibaaziz500 2 жыл бұрын

Hello , Thanks for video , it is well explained and easy to follow. I have one question , i tried to create my own dataset , i created the model , but i want to evaluate the model , i get 0 in accuracy and other parameters . Can you please help me out , i have an university project and i am stuck on this step. Thank you so much

@tylerstyle307 Жыл бұрын

i have the same problem where the accuracy is ridiculously low. If anyone knows how to fix this, please do tell

@aryandalal5409 Жыл бұрын

@@tylerstyle307 I dont know the exact reason why the accuracy is so low but you can always try different model. Models learn different parameters of the dataset to give accuracy using a different model might help it.

@joshblenkinsop6157 Жыл бұрын

Spent 2 days trying to upload my own dataset to train the model. Will not work, strongly suggest starting with another tutorial.

@ptsd7403 Жыл бұрын

Did you find any solution or alternative

@__yusuf__ Жыл бұрын

Also, the format of the images must be jpg. The problem may be here. I encountered this problem. You can convert formats such as png from some sites on the internet to jpg and upload them to collab.

@Niwly Жыл бұрын

Good day! Do you have an updated version of the collab? I'm trying to train right now but I am running out of memory installing -q tflite-model-maker. I guess they updated maybe. Thanks!

@nabeelmalik.4714 Жыл бұрын

Same problem, I've execute commands but it took very long time. Do you find any solution for this??😊

@Niwly Жыл бұрын

@@nabeelmalik.4714 Use the command Palette in colab with ctrl+shift+P. Then type "use fallback runtime version" and select. This will roll back to Python 3.9.16 runtime and tflite-model-maker can be installed. You need to connect to a runtime first, then the option will be available and it will restart with 3.9.16 instead of 3.10. Also just a heads up that apparently this will only be available until mid-May so a fix for 3.10 probably is still needed to work with Colab after that.

@nabeelmalik.4714 Жыл бұрын

No, i don't have updated version and my work is running perfectly fine using fallback version.

@vikramm4967 Жыл бұрын

@@Niwly Which will be available till mid-may? The fix which you have mentioned?

@Niwly Жыл бұрын

@@vikramm4967 yes. its not working right now

@rs45_ 2 жыл бұрын

What should be the averages precision score for module to be consider as good?

@Lolatyou332 2 жыл бұрын

Define good... If you are planning at looking at single images, then you will need 90%+ accuracy. Models used for video can be less accurate because you will have more frames to reference to attempt to find the object, but you would want much fewer false positives than false negatives. Just have to use math / statistics to determine how useful your model actually is given in the environment or context that you expect it to be used.

@II__zeus__II 2 жыл бұрын

Can you train an instance segmentation model using DeepLab v3 using the same framework and transfer learning?

@khanhleviet5416 2 жыл бұрын

Model Maker doesn't support image segmentation yet. However, you can use the deeplabv3 code in tensorflow/models GitHub repository to train a custom model. github.com/tensorflow/models/tree/master/research/deeplab

@fabriziotringali4915 2 жыл бұрын

Hi, can I also use google teachable machines to train my own trained model.tflite (tensorflowlite model)? Do you have a tutorial about it? Thanks a lot for your support.

@marcusyungchong7944 Жыл бұрын

Hi, do you have any update for this? I'm currently looking at the same thing and appreciate if there's any update from you

@ManuelHernandez-zq5em 2 жыл бұрын

I noticed the "train" folder has the jpg and xml files (the xml files were surely generated by labelImg after labeling the images). However, I don't know why the "validate" folder also has xml files. Were these xml files also generated with labelImg and with what purpose if in my opinion this folder should only have the jpg files. I mean, shouldn't the "validate" folder only contain the images files to validate (test) the generated android.tflite model?

@ethennichols5649 2 жыл бұрын

Yes the validate folder needs the xml files because it needs to what is in the image to is in order to see if the models prediction was correct

@grenadeasc Жыл бұрын

if you don't give the code the answer in the form of an xml you would never know if he did it correctly so when validating you basically do prediction(image) == answer in xml

@thupilcharwaka Жыл бұрын

ERROR: Could not install packages due to an OSError: [Errno 28] No space left on device

@bandara-sk3rf 9 ай бұрын

have you fixed this. i am facing the same issue

@kaveeshamadhubhashana3094 2 жыл бұрын

how can i move this detection into real time

@einstein_123 Жыл бұрын

while copying the model i am getting errors like "missing destination operand after /home/pi/Downloads/android.tflite" how can i fix it .can anyone help me please .it is emergency for me.

@ayeshasiddiqui9068 Жыл бұрын

While running the code I am getting no module tflite _support and not able to install it

@sanjuk2325 2 жыл бұрын

how to save checkpoint in between each epochs and retain them after and continue ?...

@defyphysics2955 25 күн бұрын

same i need this...

@bushesbakedbeans1fan888 2 жыл бұрын

Hey thanks a ton for making this guide. Currently I have a slight issue with it though, and that is at step three I get the following error when trying to use my own dataset. Here it is, (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). Can someone please help me with this?

@KingTine724 2 жыл бұрын

HEY, I am also getting the same problem! did you figure out the solution

@bushesbakedbeans1fan888 2 жыл бұрын

@@KingTine724 Nope, I just decided to use teachable machine image classification and the tensorflows GitHub's example app.

@KingTine724 2 жыл бұрын

@@bushesbakedbeans1fan888 So did it work and if it did , can you give links to the code that you used and other resources pertaining to model that you have created.

@bushesbakedbeans1fan888 2 жыл бұрын

Adafruit teachable machine guide should also work

@bushesbakedbeans1fan888 2 жыл бұрын

@@KingTine724 Okay it seems like my comments with links are getting deleted, so I'd recommend you use the adafruit teachable machine guide. I followed a different one but I think it was taken down. Another option I'm looking into is training with one of tensorflow's image classification google collabs.

@nanaasanteful Ай бұрын

Every step has been followed but keep getting this error "RuntimeError: Input tensor has type kTfLiteFloat32: it requires specifying NormalizationOpt ions metadata to preprocess input images." Raspberry Pi 4 Bullseye, 32-bit

@darkalden Жыл бұрын

tflite-model-make is completely broken and almost impossible to install. All the ways I try generate version incompatibility errors and an etc of other errors. Any easy way to install it on this date or any guide to the versions of the dependencies that will probably be installed?

@andreserafim1256 9 ай бұрын

Did you find a solution for this? Having the same issue...

@vladyslavblazhenko7569 Жыл бұрын

can i run it on windows instead of raspberry pi ?

@deepalisharma1327 2 жыл бұрын

What is that TEMP_FILE under 'Run object detection and show the detection results' section?

@kaysong3800 2 жыл бұрын

I am also curious about that, did you get any idea?

@mkeremyucedag 2 жыл бұрын

Hello, I have updated tflite-runtime to 2.7.0 but I still get segmentation error. Any recommendations?

@mkeremyucedag 2 жыл бұрын

If anyone has issued same problem, I've solved it with updating pip version then updating tflite-runtime

@kiriobalbous8210 Жыл бұрын

Hi, I'm an engineering student and I have to build a beer bot pong. I've use your video for create my model and use it. I can not return the position of my object. i've read lots of article aboute bounding box but i can not use it. does anybody know how a can return contact details? thanks in advance

@aqilhakim3459 7 ай бұрын

Sir please this is urgent when i upload the file xml it is always getting fail because my file is over 200 picture in it so i overcome this problem

@aayushsinha9620 Жыл бұрын

# Load the TFLite model ---> 17 options = ObjectDetectorOptions( 18 num_threads=4, 19 score_threshold=DETECTION_THRESHOLD, NameError: name 'ObjectDetectorOptions' is not defined I have trained the model using my dataset , but when I try to load the trained model using a image-url , it shows this error.

@MrSANS 2 жыл бұрын

Hello, how about custom object detection for react-native with tensorflow.js ?

@khanhleviet5416 2 жыл бұрын

You can use TFLite Task Library for TF.js to run the TFLite model trained with Model Maker using TF.js. Check out the example here: js.tensorflow.org/api_tasks/0.0.1-alpha.8/#class:ODCustomModelTFLite

@truonglaminh3519 2 жыл бұрын

@@khanhleviet5416 I got an issue when using TFLite Task Library. Could you let me know what happen with it? Error: Failed to create ObjectDetector: INVALID_ARGUMENT: Output tensor at index 0 is expected to have 3 dimensions, found 2. [tflite::support::TfLiteSupportStatus='400']

@khanhleviet5416 2 жыл бұрын

@@truonglaminh3519 It sounds like your TFLite model is invalid. Could you post your TFLite model on the TensorFlow forum and details about how you created that and tag me (khanhlvg@) there? I'll take a look. discuss.tensorflow.org/

@truonglaminh3519 2 жыл бұрын

@@khanhleviet5416 I have used your model in this video and it still error const model = await tfTask.ObjectDetection.CustomModel.TFLite.load({ model: 'android.tflite', }); // Run inference on an image. const img = document.querySelector('img'); const result = await model.predict(img); console.log(result.objects); // Clean up. model.cleanUp();

@Bebo991_G Ай бұрын

i tried to to use it but the pip install tflite-model-maker, downloaded about 50GB and reached google's storage limit

@ch1ldr3n27 Ай бұрын

same for me :(

@stu7075 2 жыл бұрын

is this can work on windows ??

@hung1853 9 ай бұрын

Now, I can not install tflite model maker on gg colab . It download very file and use very large memory.

@feudGame 9 ай бұрын

You can do the same locally. there are instructions there.

@hung1853 9 ай бұрын

@@feudGameit's like ssd mobilenet?

@full_bearfull_bear4783 Жыл бұрын

Why is it like impossible to find out how to load the model and continue training it? The only commands left out that everybody will need to use.

@peppiler 2 жыл бұрын

Hello, when running "python detect.py --model android.tflite" I get "Segmentation fault" Any idea how to overcome this?

@khanhleviet5416 2 жыл бұрын

I think you're using an older version of tflite-runtime (2.5.0). Please update to the latest version (2.7.0) by running "pip install tflite-runtime --upgrade"

@peppiler 2 жыл бұрын

@@khanhleviet5416 It works, thank you!

@nguyenhoangphuc98 11 ай бұрын

What do you do upload your dataset folder to Google Colab ?

@TensorFlow 10 ай бұрын

You can either upload the files directly to Colab runtime through the upload button on the left hand side, or link the Colab session with Google Drive. Check out this notebook for more options about uploading data to Colab: goo.gle/3qxoIOq

@pakpomsang5 2 жыл бұрын

Sorry, I have one question, can I use this model with Object Detection on android studio

@elmehdibaaziz500 2 жыл бұрын

yes

@agroconsultant7340 Жыл бұрын

helllo. can i chenge category names inn colab?

@leventtoprak8250 Ай бұрын

I am getting error while pip commands executing. is there any solution? I bought pro version of colab, before that disk space was not enough.

@user-oi3vx2ul1f Ай бұрын

Everything works better with 3.8 python and you have to resolve pip installation process yourself. It related with libraries dependencies you have to install in your system and finally CUDA Nvidia packet of libs to make tensorflow works correctly. I just installed yesterday. Now everything works good and I had many errors with dependencies.

@xxegyzz5250 2 жыл бұрын

I was trying to follow your steps and I'm half way through but I am getting an error 'int' object has no attribute 'label_map' in the training part. What should I do? This is my first time customizing my own dataset I am having a hard time figuring it out. Please help.

@khanhleviet5416 2 жыл бұрын

It's hard to troubleshoot code issues on KZfaq. Can you post a notebook to reproduce your issue to the TensorFlow Forum discuss.tensorflow.org/ and tag me (khanhlvg@). I'll take a look.

@apijaysharma Жыл бұрын

Same here did you got any solution? I found one but wasn't helpful it said to install numpy version 1.11 bt I couldn't install

@xxegyzz5250 Жыл бұрын

@@apijaysharma No, I did not got any solution so I used edge impulse instead. It is very easy to follow and use. Sorry for the very late reply btw.

@ruchikmahesh755 2 жыл бұрын

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. This error is not getting right what ever I do...Please it would be great if anyone told me how to fix this error as I am working on a project and need this to make my object detection work!Thanks in advance..

@rokaskirdulis8458 2 жыл бұрын

Hi, I am having the same problem, I you have already fixed this error, please tell me what is wrong. Thanks

@castheeuwes1085 Жыл бұрын

Same here. Did anybody solve this?

@tomplatner8890 Жыл бұрын

Also having the same problem....Somebody help!

@kevinyoung9243 2 жыл бұрын

anyone know what that image labeler at 3:15 is

@elmehdibaaziz500 2 жыл бұрын

LabelImg

@muhammadshayan8794 2 жыл бұрын

Great. How to execute it on Pycharm?

@avinashsoni9533 Жыл бұрын

how you make .xml file with jpg please suggest

@jhonvargashuarcaya9106 Жыл бұрын

LabelImg

@user-cv4jm2sz5k 2 жыл бұрын

10:41(python3 detect.py --model android.tflite) -> segmentation fault😢😢😢 plz help me but python3 detect.py --model efficientdet_lite0.tflite work well

@mugtabataha3355 2 жыл бұрын

I have problem with downloading the zip file from the drive can you tell how you passed that step😅

@user-cv4jm2sz5k 2 жыл бұрын

@@mugtabataha3355 tflite file? I think Just drag or right mouse click -> download

@mugtabataha3355 2 жыл бұрын

@@user-cv4jm2sz5k I meant the file that contains your own dataset😅

@user-cv4jm2sz5k 2 жыл бұрын

@@mugtabataha3355code Comment is automatically deleted by youtube that happened to me too, I think that happened when the data loading from the drive to colab failed.

@mugtabataha3355 2 жыл бұрын

@@user-cv4jm2sz5k thank you so much :) I wish you all the best with your project

@luizabbadia3411 Жыл бұрын

Please, I receive a bunch of errors : Building wheel for fire (setup.py) ... done ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. xarray 2022.12.0 requires packaging>=21.3, but you have packaging 20.9 which is incompatible. statsmodels 0.13.5 requires packaging>=21.3, but you have packaging 20.9 which is incompatible. plotnine 0.10.1 requires matplotlib>=3.5.0, but you have matplotlib 3.4.3 which is incompatible. mizani 0.8.1 requires matplotlib>=3.5.0, but you have matplotlib 3.4.3 which is incompatible. What is going wrong??? It was working OK...

@luizabbadia3411 Жыл бұрын

Well I think I solved! Removed the '-nightly' parameter

@karimabdalla3136 Жыл бұрын

@@luizabbadia3411 I'm getting ERROR: Cannot install tflite-model-maker==0.1.2, tflite-model-maker==0.2.0, tflite-model-maker==0.2.1, tflite-model-maker==0.2.2, tflite-model-maker==0.2.3, tflite-model-maker==0.2.4, tflite-model-maker==0.2.5, tflite-model-maker==0.3.0, tflite-model-maker==0.3.1, tflite-model-maker==0.3.2, tflite-model-maker==0.3.3, tflite-model-maker==0.3.4, tflite-model-maker==0.4.0, tflite-model-maker==0.4.1 and tflite-model-maker==0.4.2 because these package versions have conflicting dependencies. after running "!pip install tflite-model-maker", please help

@seanmuirhead2156 Жыл бұрын

did it take anyone else 10 years to download dependencies?

@DSET_ChiragNRaj 11 ай бұрын

yes for me🤣

@TechCree 2 жыл бұрын

Without the one list of data that the object recognition generates, it's all just a nice gimmick that doesn't help. has someone managed to create a list? nothing comes from the Tensorflow team,

@prashantsharmastunning 2 жыл бұрын

which raspberry pi is that?

@levietgiakhanh9818 2 жыл бұрын

It's a Raspberry Pi 4 with 4GB RAM.

@ananditachakraborty7439 Жыл бұрын

I guess this tutorial gon obsolete

@user-jl8dm5vq5j 8 ай бұрын

USELESS VIDEO CURRENTLY....takes almost eternity just to download the necessary packages....additionally not enough space too...

@user-uf9ky8pq5s 11 ай бұрын

I have a joke on google, but it is deprecated.

@abdullahimtiaz6564 2 жыл бұрын

When I import model_spec and quantizationConfig from tflite-model-maker it gives error notfounderror: /usr/local/lib/python3.7/dist-packages/scann/scann_ops/cc/_scann_ops.so: undefined symbol: (a random code) I tried everything that was available and recommended to do so but couldn't resolve the error. Why it's generating error on colab ? It was working fine a week ago but now error occurs everytime