TFOD 2.0 Custom Object Detection Step By Step Tutorial

Google colab code
colab.research...
Files of Object Detection
drive.google.c...
⭐ Kite is a free AI-powered coding assistant that will help you code faster and smarter. The Kite plugin integrates with all the top editors and IDEs to give you smart completions and documentation while you’re typing. I've been using Kite for a few months and I love it! www.kite.com/g...
All Playlist In My channel
Interview Playlist: • Machine Learning Inter...
Complete DL Playlist: • Complete Road Map To P...
Julia Playlist: • Tutorial 1- Introducti...
Complete ML Playlist : • Complete Machine Learn...
Complete NLP Playlist: • Natural Language Proce...
Docker End To End Implementation: • Docker End to End Impl...
Live stream Playlist: • Pytorch
Machine Learning Pipelines: • Docker End to End Impl...
Pytorch Playlist: • Pytorch
Feature Engineering : • Feature Engineering
Live Projects : • Live Projects
Kaggle competition : • Kaggle Competitions
Mongodb with Python : • MongoDb with Python
MySQL With Python : • MYSQL Database With Py...
Deployment Architectures: • Deployment Architectur...
Amazon sagemaker : • Amazon SageMaker
Please donate if you want to support the channel through GPay UPID,
Gpay: krishnaik06@okicici
Telegram link: t.me/joinchat/...
Please join as a member in my channel to get additional benefits like materials in Data Science, live streaming for Members and many more
/ @krishnaik06
Connect with me here:
Twitter: / krishnaik06
Facebook: / krishnaik06
instagram: / krishnaik06

[@pranilpatil4109]

I had been stuck in training for 2 weeks and now that I found your video, such relief that I am able to train finally.

[@karanbagga4665]

I'm stucking at training and it don't goes at 100th step what i do now?

[@pranilpatil4109]

@@karanbagga4665 whats error?

[@lukasznowarkiewicz]

If you are trying to reproduce this tutorial in November 2023 with tf 2.14 this may be solution to you as well: !pip install tensorflow[and-cuda]==2.13.0

[@oguzhancoskun9047]

Thanks a lot mate

[@vishakarudhra8665]

This was a rapid and fantastic journey, Krish sir. I would have loved to this by myself but I am surely not ashamed to say I followed your tutorial to build this model. Thanks sir. You are a gem.

[@apoorva3635]

10:31 - The last 4 files in the video do not match the ones present in the Google Drive link.

[@sherafati]

This is the best Tensor flow object detection api tutorial! Explained so clearly

[@sathyanirmanifernando]

@Krish Naik , There is no object_detection.py file in the drive. could you add ??

[@anushadatta463]

hi, did you manage to find it?

[@ruhulhaque3407]

@@anushadatta463 No I couldn't. Did someone get this object_detection.py file in the drive

[@thilaksk2171]

Hello did u get that object_detec.py file ?

[@sidharthanand4284]

did you get it?

[@sidharthanand4284]

@@thilaksk2171 any updates?

[@smitsavaliya]

i was finding this tutorial last few days and when i see it, just perfectly detailed information and easy to understand without any major reading of full documentation from tensorflow. So basically thank you

[@ceng5016]

Hello Mr. savaliya. I have tried for 1 week on google colab. But ı guess I have a problem about packages versions. Please can you tell me your environment and versions vs. Thanks for your corporation

[@omarhernandezcalvario8256]

Hello! Where is the file object_detection.py?

[@farhanrashid786]

@@omarhernandezcalvario8256 did you find that file..i also need that file

[@vishnuvahini127]

@@farhanrashid786 did you find the object_detection.py file?

[@farhanrashid786]

@@vishnuvahini127 nope

[@mohammedkaifmirza7585]

hey guys, I would suggest you to change batch_size=2 from 8 if the training is not working in this case. I was facing this issue for hours, Then i come at this conclusion to reduce the batch-size. Hope it will work for you also.

[@eswaramoorthyk6312]

For those of u trying it out post 2022, "tensorflow-gpu package is removed", It is enough to install "tensorflow" library alonw

[@dhruvekiyawat1567]

why this is so

[@letslovestraydogs4648]

do other have problem with running the code on tensorflow-gpu 2.8 ? for me training gets stucked

[@shubhankarvashishth2755]

Thanks a lot Sir !!! I was waiting for it since ages. Waiting for more. Pls post more, I am on a learning spree.

[@felixjarwis3763]

Node: 'ResNet101V1_FPN/model/conv1_conv/Conv2D' DNN library is not found. [[{{node ResNet101V1_FPN/model/conv1_conv/Conv2D}}]] [Op:__inference__dummy_computation_fn_20479] Can not find solution for this mistake. Anybody had this problem?

[@kenzilamberto1981]

I don't find the Object_detection.py file in the drive that you shared, do we need it? any other way of obtaining it?

[@aishwaryajois7923]

did you find it?

[@deveshverma6658]

Hi Krish .. Object Detection file is missing .. .from where can i get this.???????????

[@devathisailokesh8644]

Yes it is missing

[@ayomidarshika4693]

How can we get that file

[@YadavSachin01]

Which time use XML file images classification timing please tell me

[@asmitagulati7950]

I am getting version errors in the Cudnn library, please help. It is showing that loaded runtime CuDNN library: 8.0.5 but the source was compiled with: 8.1.0.

[@coderforchange418]

did you find answer ?

[@asishissacalex4148]

Can we do the training in epochs instead of number of training steps? Had to do it, so as to compare with another models.

[@abhishek_maity]

FINALLY!!! love u sir 🙂😁🔥🔥🔥

[@MelodicMirth]

can you please share the object_detection.py file as it's not available in the drive

[@nit8826]

Stucked on Use 'tf.cast' instead Please help 🥺

[@harshalsangame1305]

@@nit8826 i think because train.record and test.record do not have bytes inside. it just a empty file

[@shloksharma5629]

Running into the error: "loaded runtime cudnn library 8.0.5 but source was compiled with 8.1.0" Pls help. Thank you

[@shaonsikder556]

Is it limited to bounding box? Can we use this for polygon or point annotations? Another question, based on which criteria you choose that specific model? Is it random choice? Thanks sir in advance.

[@0xN1nja]

you can get coordinates of bbox and draw it using cv2.line (but its not recommended)

[@npathak20]

Bro Krish Naik u r a legend brother. Ultimate respect

[@vamsikrishnanadimpalli6678]

Hello Krish, I did not find the Object_detection.py file (inside training_demo folder) in the drive link given the description. Could you please tell me where I can download it from?

[@santhoshmanoharan8969]

there is a colab link in the description. check that

[@niranandkhedkar3681]

@@santhoshmanoharan8969 Nothing like object_detection.py in colab and the drive

[@shaileshnanisetty4805]

@Krish Naik ,There is no object_detection.py file in the drive. Could you please upload it, you might have forgotten.

[@rohit_mondal__]

Actually the object detection.py file,Is not a required file for training the model. Its just for using the model...which can be created easily upon browing the internet

[@abehera5974]

@@rohit_mondal__ can you please provide a link for that?

[@AkshayDudvadkar]

where is object detection file in drive

[@phoque6]

May i know where i can get the object detection python file from as it is not in the google drive? thanks

[@thilaksk2171]

Object_detection.py is missing in the folder please do update it.

[@classicemmaeasy2292]

pls have you gotten it

[@austinmark242]

Krish can you make a playlist on Tensorflow.

[@VinayKumar-mu9qu]

I can't find 'object_detection.py' file in your drive link. I'm getting an over there. Please help me with that sir @krish naik sir

[@sagnikroy6405]

Thanks, Sir !! Finally . Also, can you make any video regarding how to interpret RCNN from scratch?? Like coding with theory, that will help us surely. :D

[@RedShipsofSpainAgain]

10 advertising interruptions on this video when watching on mobile device. 0 ad interruptions when watching on laptop with AdBlock installed. Guess I won't be watching these videos on my phone anymore. But excellent content Krish, thank you!

[@vedanthbaliga7686]

Just switch to Brave browser. You wont get advertisements

[@sasikanthreddy3116]

Can we do incremental training without training old labels, Ex: I trained 5 labels and the accuracy levels are perfect and now I wanted to train two new labels and make the labels count to 7, I just don’t want to disturb the trained 5 labels. Without training old labels, can we train new ones. Is this possible using this model. Thanks for suggestions in advance.

[@rabomeister]

it does not make sense. at the deepest point, computer tries to find connections and logic, think that you have gaussian and as normal, u can predict the rest of that graph. the computer does the exact thing but with incredibly extreme connections.

[@Sebastian-hv7jz]

The code that you paste at 42:57 in your video is missing in your download section. Can I have that code please?

[@japhethmumo6768]

Thanks alot Krish, Your tutorial has helped me, I would kindly request you to do a tutorial which shows how the tensorboard can show other evaluation metric like precision and recall as currently we are only provided with loss metrics.

[@thinktwice8480]

Thank You Sir for an extremely helpful tutorial ! Can You please explain Inference Script created by you. And also do Inference script remains default ?

[@saifchowdhury9728]

Sir, needed a tuitorial video for covid 19 radiography detection with cnn and rnn.

[@anshumanmaitra4784]

In case of testing the model_builder_tf2_test.py , I am getting this attribute error saying that there is no attribute named 'experimental'... please can anyone resolve it?

[@KeerthanaSedb]

Please help I can't upload train.record and test.record file I donno why

[@umairsabir6686]

Thanks for this video. Can you please make one video on how to train a Mask RCNN using this same Tensorflow API on a custom dataset. TF API has released support for Mask RCNN in TF 2.x as well.

[@mahajanshilpa141]

Are you able to run the abovesaid model....???

[@umairsabir6686]

@@mahajanshilpa141 are you asking abt TF1 implementation or TF2 ?

[@mahajanshilpa141]

Tf2.. which has been described in the video..

[@HamzanMartinez]

Did you figure out the mask model method yet?

[@umairsabir6686]

@@HamzanMartinez I applied it using the original mask rcnn git repo but not using TF2 API

[@dicksonneoh5563]

How do I freeze the backbone of the pretrained model and just train the head? Probably that can reduce the number of training iterations for a good result?

[@mythorganizer4222]

for layer in model.layers: layer.trainable = False

[@dicksonneoh5563]

For TF object detection API the model is defined in the config.pipeline file. I am not sure how to run the python code snippet you posted

[@mythorganizer4222]

Try using hooks. Then run the code.

[@frankgkilima2510]

Thank you for the very good and well prepared tutorial. It's very helpful!

[@mrziedaan]

I think the object_detection.py is missing from your drive

[@karthika8192]

Yes

[@KhareBeats]

Please Help!!! I am getting the following error while training the model: -----------------------------------------------ERROR MESSSAGE ---------------------------------------------------------------------------------- Loaded runtime CuDNN library: 8.0.5 but source was compiled with: 8.1.0. CuDNN library needs to have matching major version and equal or higher minor version. If using a binary install, upgrade your CuDNN library. If building from sources, make sure the library loaded at runtime is compatible with the version specified during compile configuration.

[@mahajanshilpa141]

thank you so much sir.... but while training i got an error... Errors may have originated from an input operation. Input Source operations connected to node ResNet101V1_FPN/model/conv1_conv/Conv2D: In[0] ResNet101V1_FPN/model/lambda/Pad (defined at /usr/local/lib/python3.7/dist-packages/object_detection/models/keras_models/resnet_v1.py:51) In[1] ResNet101V1_FPN/model/conv1_conv/Conv2D/ReadVariableOp: Kindly help in this.... All in vain if at the end this error stops the training.

[@M.Pranathi_28]

were you able to resolve it?

[@anujsindgi5768]

Great video. You could also show how to get an mAP value for the trained/exported model.

[@reallywhoamii]

Thanks and this is a very nice piece of tutorial. However, when I started the training, I encountered an error "cannot import name '_registerMatType' from 'cv2.cv2'". My guess is some repositories updated their builds and broke something since your tutorial is already sometime back. Do you have a latest tutorial?

[@sachinpoonia4989]

Same here

[@neerajashish7042]

same

[@vikasnunna7341]

Try executing the below before training the model. It worked for me !pip install "opencv-python-headless

[@acnice_]

@@vikasnunna7341 thanks

[@prabhatsingh1964]

!apt-get -qq install -y libsm6 libxext6 && pip install -q -U opencv-python use this command to install open cv

[@vishnuvasansrinivasan7797]

May I know where the 'object_detection.py' can be found in the drive link that was shared with us? If not where can that be found?

[@nit8826]

Stucked on Use 'tf.cast' instead Please help 🥺

[@Artistic_abhishek]

there are only 3 files uploaded where is the 4th file

[@chinmayganguly457]

the object_detection.py file isn't there in your zip file, the other 3 are there though

[@abdelhamidazanzal4403]

hi! i notice the same thing. where did you find the object_detection.py file if you have already find the answer

[@wtfJonKnowNothing]

If you're using Colab, setting up all this lengthy installation for just 12 hrs session is like hell.

[@shardulkhadye5875]

setup once in your google drive , then u just have to mount ur google drive on colab and work from there

[@granatapfel6661]

@@shardulkhadye5875 can u explain how?

[@user-zc2uh2zg2n]

sir there is no file (google drive) of object detection inside the training demo

[@allea-zb7kl]

From the training process, can we see the weight and bias results for each layer that will be used for test data?

[@kishorekumar6984]

Hey Krish, can you upload the object detection .py?

[@ritikvaidande4683]

28:07 dragging the file works instead of downloading and then uploading :)

[@sainathpawar7248]

I am facing this issue while training. Node: 'ResNet101V1_FPN/model/conv1_conv/Conv2D' DNN library is not found. [[{{node ResNet101V1_FPN/model/conv1_conv/Conv2D}}]] [Op:__inference__dummy_computation_fn_24455] Any quick solution to this

[@TheZipzapzum]

I am facing same issue .. did you fixed it ?

[@sainathpawar7248]

@@TheZipzapzum Not able to fixed this issue, but you can refer below youtube link for object detection. I am able to executed steps successfully. kzbin.info/www/bejne/l564g6yJa6h0l7s

[@prernanichani477]

Finally !! Thankyou Sir. You made it so easy. Also, a request can you please record for using the same trained model in local system for prediction.

[@kanchanadevim3231]

There is no object detection (python file) file in training_demo... Urgent Plz send that file sir.

[@techprinvid338]

Hi sir , Should i need work station or best laptop to create datascience projects ?

[@metaLeap]

Hi Sir, I followed all the steps as instructed, but during the training step, its freezing on the line: Use fn_output_signature instead A couple of times it even stopped(^C). What can be the reason for it?

[@saibalmaiti3581]

Yes I also got that, did you find any solution?

[@MVEC_SaiChakradhar]

@@saibalmaiti3581 Reduce the batch size to 4 or 2

[@prabhatsingh1964]

reduce the batch size to 4 and it will work

[@user-zh9ot4tz9c]

i follow all these steps and also run your code but get this error AttributeError: module 'tensorflow.python.ops.control_flow_ops' has no attribute 'case'. i am trying to make my final year project , and face this issue from 3 months but get no solution.kindly help me

[@charansammeta3405]

did you resolve?

[@gangaramyadav6104]

Wow really great sir 👏 thanks alot for this wonderful content..

[@prashanth_b_k]

The object detection file is missing , where can I download that??

[@sethvishu]

Getting this error after training : Errors may have originated from an input operation. Input Source operations connected to node ResNet101V1_FPN/model/conv1_conv/Conv2D: ResNet101V1_FPN/model/lambda/Pad (defined at /local/lib/python3.7/dist-packages/object_detection/models/keras_models/resnet_v1.py:51) Function call stack: _dummy_computation_fn

[@azgiraffe]

Same

[@adenijidamilola2910]

Hi Krish, thank you for the tutorial. I was following the tutorial and everything was going fine until I wanted to train the model, I was encountering this issue ***Input of GeneratorDatasetOp::Dataset will not be optimized because the dataset does not implement the AsGraphDefInternal() method needed to apply optimizations.*** Is there any way I can skip this part and make it work. Thank you

[@mehreenkhan7356]

Sir kindly upload a video about brain tumour segmentation by using deep learning with U-Net architecture

[@jk9999999able]

Hey great tutorial! i have 2 questions: i have tensorflow 2.7 installed is it okay? also the function -m pip install takes forever and crashes my collab after 1h. I dont know how to deal with this

[@syedamisbah6534]

Sir kindly upload a vedio about sequential minimal optimisation using svm

[@shanmugamramanathan6021]

object_detection.py file is missing in your gdrive

[@salmansolaiman4260]

Sir? the object detection in google drive is missing?

[@rajkir2852]

Object detection : classified + localized

[@abhisekdatta8887]

should be multi labeled image

[@kartikec1752]

Nice work! thank you for sharing.

[@otitio4175]

how to do test in anaconda prompt its showing could not find conda environment test

[@kratos_gaming9765]

broooooooo woh object_detection wala file nai hai apke drive mai

[@ShubhamKhan]

*Sir, tell me how I get the accuracy, with confusion matrix, precision, and recall in this model.*

[@classicemmaeasy2292]

from sklearn.metrics import confusion matrix

[@sreechaitanya1178]

when I am downloading ResNet 101 FPN in colab it is showing "Cannot write to ‘ssd_resnet101_v1_fpn_640x640_coco17_tpu-8.tar.gz’ (Read-only file system)" like this. Is there another method to do extraction?

[@niksvp93]

You have to load it in the same directory.

[@diseaseddates4242]

amazing content really appreciate it. Question the running of training model is stuck on Instructions for updating: Use `tf.cast` instead. what could be the reason ?

[@patanahi4873]

did you find any solution?

[@nit8826]

@@patanahi4873 same stuck in Use 'tf.cast' from 5 hours can you please help me

[@harshalsangame1305]

@@nit8826 i think because train.record and test.record do not have bytes inside. it just a empty file

[@kranthikumar4674]

really awesome sir, much needed video for students who really wanted to do the real-time project with their own dataset. Sir, also make me a video on OCR models.

[@user-xy4vz2uj9m]

Thank you for video. It is very helpful! I have a question about the model config file. In that file, there are some lines like eval_config, and eval_input_reader. Those eval means validation during the train? or test the model?

[@sarveshkumaryadav6222]

Sir, will you launch new full live course on data science

[@rajkir2852]

Well done sir

[@dheeraj4761]

Hey. Can anyone suggest a command to retrieve performance parameters of the model? Thanks.

[@anitannajoseph7796]

Sir, can you make a video regarding demand prediction in business?

[@user-tz6nn8iw9m]

Thank you, your video helped me a lot :)

[@BrownCookieBoy]

Got some error when training the model, had to do with cudnn version. I basically did this all over again (not sure if i had to start from scratch) but with tensorflow 2.4 instead using "!pip install tensorflow==2.4.0"

[@crix177]

tensorflow.python.framework.errors_impl.UnknownError: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above. this error?

[@Lifetree_evolution]

@@crix177 yes, did you find the solution ?

[@crix177]

@@Lifetree_evolution yea, change tensorflow version to 2.4.1

[@azgiraffe]

Hi how did you test the model? There was no object_detection.py

[@nguaial8490]

@@crix177 !pip install tensorflow==2.4.1 or !pip install tensorflow-gpu-=2.4.1? Which command did you use? Krish's video only installed tensorflow-gpu.

[@GuessMeable]

Hello, I'm not getting any steps after running model_main_tf2.py, its getting stopped after few minutes without giving any specific issues. Anyone facing similar issue ?

[@dinugakasun5218]

This is superb, please do custom yolo object detection video too. Thanks!

[@Student_yet]

sir its a request to please make one video on using YOLOv8 also.

[@edgarramos4830]

Does anyone had the following error while training the model? "tensorflow.python.framework.errors_impl.AlreadyExistsError: File system for s3 already registered"

[@akshayakki5692]

Hello sir, can you please tell me how to load the tensorboard to the trained model for evaluating the performance of the model.

[@abhisekdatta8887]

While training the pipeline I got this: "Instructions for updating: Use fn_output_signature instead" I am using google colab with GPU and 60% usage of the memory. How to solve sir ...

[@sampathvinaykumarreddymajj790]

Issue solved???

[@abhisekdatta8887]

@@sampathvinaykumarreddymajj790 yes bhai .. Start with a fresh colab

[@sampathvinaykumarreddymajj790]

@@abhisekdatta8887 Thank You bhaii

[@andrei4221]

thanks for the video! I liked!

[@amanroland1438]

No one explains evaluation. i have trained a model in colab but when i run evaluation it only takes the last or latest checkpoint to evaluate. how do i configure it to run along with training. or use all checkpoints.

[@ChatMan011]

Sir.. training not happening at the end...just does something warning.. warning....and then loss is not appearing..can u help me

[@DACSAmanpreetSingh-eh7gt]

facing the same issue.

[@DACSAmanpreetSingh-eh7gt]

if you find a solution do ping me

[@metaLeap]

I believe this is due to the ram usage restrictions of google Colab, try reducing the batch size more and see if it works. I also faced this problem and eventually shifted to training on my local system (cpu) by creating an Anaconda virtual environment. Although it's very time consuming , it does work. If you have the patience to wait for hours before seeing the first 100 steps results then I'd recommend you train on your local system. Also in case anyone finds a solution to the Colab training issue pls do reply

[@ojaskulkarni6038]

How do I save the whole folder to google drive as I am losing all my data and files once colab gets refreshed. Any solution to this?

[@chaitanyaawasthi7711]

1) mount your colab to google drive, drag on drop every local file to " your drive " section

[@jayasandeepreddy3474]

Hi Krish image height and width should be maintained same for all images in train folder or it may vary ? If it is varied how to deal with them ?

[@luigiesposito2481]

hi bro great video and description, i wanted to ask you i don't understand much, i wanted to ask you but is it possible to change class and put people? and use it in pc monitor screenshot shot? that is, in real time?

[@codingphilosophy5497]

How can we build our own backbone model to embed in this tensorflow object detection api? And which files we have to update for this purpose? Let me allow to explain my question, let suppose I introduce my backbone model "A". In which files I have to give link of "A"? and Where I have to place "A.py" in object detection api?

[@anuraggoyar]

Nicely explained, Thanks 😊

[@felixfoo5652]

Hi sir, for this error Input of GeneratorDatasetOp::Dataset will not be optimized because the dataset does not implement the AsGraphDefInternal() method needed to apply optimizations. how to fix this error, i stuck at the training part which is after the tf.cast there

[@help2office427]

Sir, what's happening with RL Playlist

[@aizalorenzana7349]

Finally great video

[@harshalsangame1305]

After exporting my trained Model iam getting this error While Testing with image1.jpg ImportError: cannot import name 'string_int_label_map_pb2' from 'object_detection.protos'