YOLOv3 in the CLOUD : Install and Train Custom Object Detector (FREE GPU)
ฝัง
- เผยแพร่เมื่อ 2 มิ.ย. 2024
- Learn how get YOLOv3 object detection running in the cloud with Google Colab. Walk-through the steps to run yolov3 with darknet detections in the cloud and how to train your very own custom object detector. ALL WITH FREE GPU! This tutorial covers it all!
#yolov3 #objectdetection #cloud
THE GOOGLE COLAB NOTEBOOK: colab.research.google.com/dri...
This video cover:
1. Setting up Google Colab as a cloud VM with Free GPU.
2. Commands to get Darknet with YOLOv3 weights installed and running.
3. YOLOv3 pretrained coco model detections in the Cloud.
4. Configuration for Custom YOLOv3 Training in the Cloud.
5. Training Custom YOLOv3 Object Detector in the Cloud.
-----------------Resources------------------
Github: github.com/theAIGuysCode/YOLO...
Generate Open Images Custom Dataset (recommended): • Creating a YOLOv3 Cust...
Create Dataset with Manual Annotations: • Create Labels and Anno...
Train YOLOv3 on Local Machine (configure training): • Train YOLOv3 Custom Ob...
If you enjoyed the video, toss it a like! 👍
To Subscribe: / @theaiguy
Thanks so much for watching!
- The AI Guy - วิทยาศาสตร์และเทคโนโลยี
Let me know what classes you are making your custom detector detect! Also comment if you have any issues!
thanks man great video . But I have issues in installing the gpu version in my laptop . While using cmake configuration, cmake won't detect my cuda drivers . I have installed cuda 10.2 Also i didn't install cudnn
@@mubinmodi7529 for this video you don't need cmake and cuda will come predownloaded on the cloud VM. Did you mean to comment on one of my different videos?
it's saving weights after every 100 epoch. how can we override this default behavior and save weights after let's say 1000 epochs.??
Good afternon sir, I have one more suggestion kindly make a video on how to install and use detectron
Manu Tyagi awesome will do!
Your in depth explanation solved so many problems! Thanks AI Guy! I wish this video had more views!
Appreciate the kind words and glad you liked the video! Thanks so much!
Exactly
Thank you for your hard work to frame format the content beautifully and to make it work :)
Hey, great tutorial dude! I spent hours figuring all of this out and got it running back in Nov 2019, this video is a lifesaver for anyone who's new to YOLO! Appreciate the efforts man, cheers!
Done everything from scratch with the help of you, really informative and god tier tutorial :D. Thanks!
god tier indeed.
Thank you so much for this beautiful written Google Colab notebook. It is very helpful. This channel should get a lot of subscribers. Keep it up :)
Thanks so much!
10:57 was so creative that you earned my subscription
Appreciate it. Thanks a lot! :)
I followed all your tutorial on YOLOv3 and within 2 days i was able to setup YOLOv3, detect object with pre-trained model, create datasets, train them and i was able to detect custom object. Thank you for you tutorials.
Thanks a lot for the video, I was having a lot of trouble installing DarkNet and cuda in my computer and all that stuff, so this was really usefull. Great video, dude, keep it up!
Glad it worked for you! Thanks!
Thank you so much! you saved my life! hehhehehe
Awesome job!! I was struggling so much to just install TensorFlow with GPU on my local machine, and this solves everything! Thank you so much, keep iy up!
Hey man, love this tutorial on the cloud. I struggled so much getting it work on the local GPU, so glad to see a much better way!
Glad you enjoyed it! I really appreciate the kind words.
@@TheAIGuy Dude, you're killing it with great content. It's truly a great time to be alive and seeing how you can custom label objects, build a detector, and run it in real time. All for free! Would love to know you're background in tech and aspirations besides growing the channel, and openness to possible collaboration down the line.
@@argishtib Feel free to reach out to me at my email as I'd love to chat more. You can find it in my About tab of this channel.
@@TheAIGuy Sorry just saw this now. Just sent you an email.
I was hoping you will say "welcome back aliens"
Thats reserved for telusko
For all those encountering the following error in STEP 3 line 44
"
Traceback (most recent call last):
File "generate_train.py", line 4, in
os.chdir(os.path.join("data", "obj"))
FileNotFoundError: [Errno 2] No such file or directory: 'data/obj'
"
go to the generate_train.py and change the line4 from os.chdir(os.path.join("data", "obj")) to os.chdir("data")
and image_files.append("data/obj/" + filename) to image_files.append("data/" + filename)
.
It seems the images are saved in a data folder directly instead of data/obj even when you follow the steps correctly
I'm getting this error even though my images are inside darknet/data/obj
@@N-e0N Post your errors here and maybe we can help
@@leafteaneko Its okay, I tinkered around and made do. For some reason, the code really misbehaves at times for me.
I added "darkflow" to os.path.join and also made it darknet/data/obj+filename in the image_files.append. That worked for a bit but then I got an error in the training phase when it said "Can't load images from darknet/data/obj" . At this point I went back and changed everything to original and loaded it...then it worked fine!
@@leafteaneko Ok I'll take you up on your offer because the minor tweaking no longer works. Here's the error message :
Traceback (most recent call last):
File "generate_train.py", line 4, in
os.chdir(os.path.join("data","obj"))
FileNotFoundError: [Errno 2] No such file or directory: 'data/obj'
But there is a data folder inside darknet, and it does contain obj inside of which are all the images and annotations.
@@N-e0N that is really weird, the same thing happened to me once before but I did a factory reset runtime and started fresh and it worked. I know it works because I successfully trained my model and I have the config files too. Drop me a email so I can send over my code, maybe that'll help your case(?)
This is awesome, thanks a lot! Couldn't get darknet to work on my own PC (cuda and opencv troubles) for 2 days, and this solved my suffering. Took a bit of puzzling to get it to work for Yolov4, but totally worth it.
Hi Mr AI Guy. Just wanted to say a million thanks for your help working through issues I had and for creating such a great tutorial. I have my object detection via jpeg and now I can move on to getting object detection working on a live webcam. Like and subscribed. Thanks again.
Glad you got it all working!
Bro make video on predicting yolov3 on video and exporting the video in windows
The video was really helpful! Thank you so much for your effort! Could you please give me an idea how to apply the model on multiple test images (I have test.txt file analogue to the train.txt file, containing the path to the image) and create a csv file (or pandas dataframe) with the bounding box coordinates, the center (x,y) and the class probability?
You can check out the readme for alexeyab/darknet github repository, he has amazing documentation and shows how to test on multiple images at once!
@@TheAIGuy I tried multiple things that alexeyab suggests. The following one is running without an error !./darknet detector test data/obj.data cfg/yolo-obj.cfg yolo-obj_final.weights -dont_show -ext_output /mydrive/yolo/images/imgname.jpg result.txt but the result.txt is not saved in darknet repository and I cannot find it. Did you had the same issue?
@@hristinabiserinska2407 That will mostly likely save the results.txt into the colab VM. Try making it /mydrive/yolo/results.txt or something similar!
You sir are one amazing man. This video deserves more views.
What can I say? You're the best! All your videos are freaking awesome!
I followed every steps of your tutorial but at the when i start custom training it gives me the error "/bin/bash: ./darknet: Is a directory" .Please help me in this
Were u able to solve this?
I have the same issue
some how after trying many solution i get one go to darknet directory by "cd darknet" and after that run this command "!chmod +x darknet" this will give you permissions to execute the darknet files after that run your custom detector hopefully it works
I'm having same problem. "!chmod +x darknet" doesn't solve for me. AI Guy, can you help us?
hey guys i solved this problem. at some point i was in the wrong directory so my data/obj was is /content/ instead of content/darknet/data/obj. my yolov3_custom.cfg was also not appearing in the correct location. i ended up factory reseting the runtime (in the runtime menu) and then started again taking more care to make sure everything was popping up in the corrent folders and checking up current directory with %pwd if i wasnt sure. AI Guy thanks heaps for these tutorials. keep them coming!
Traceback (most recent call last):
File "generate_train.py", line 4, in
os.chdir(os.path.join("data", "obj"))
FileNotFoundError: [Errno 2] No such file or directory: 'data/obj'
I have also done %cd content darknet as just %cd darknet gives no directory named darknet. I have obj.names and obj.data in my data folder.
I have tried everything. Can someone help me out? Do I need to change something in the code?
I have a same issue here
I found the problem, you have uploaded the obj.zip and extracted it outside darknet/data and in another folder named data. Make sure that you extract it in darknet/data otherwise it will not work
# unzip the zip file and its contents should now be in /darknet/data/obj
!unzip ../obj.zip d darknet/data/obj
Wow! I have my fingers crossed, but so far this is much easier than running YOLOv3 with Darknet and OpenCV on Windows 10 for me :-) Thank you for the great tutorials!
Peregrinus Solutions LLC Thanks for the support! Glad you enjoy.
you da man! i literally struggled to do this but your tutorial helped me a shit lot. Liked and subbed!
We need video about yolo detection in webcam realtime feed.. Thanks in advance :)
Will do!
@@TheAIGuy yeah i really need to do this on webcam realtime
@@aishanoorbhatti5850 check out my latest video. Its all about running yolov3 on webcam or video in real-time! It will walk you through how to do it with GPU or CPU.
@abdessamad makdad yes I will try and figure that out! In the meantime you can train your model in the cloud and then use my latest video to see how to run it on video and webcam!
@abdessamad makdad you can still run video detections on CPU! You wont get as many frames per second but it still works!
May I know how download trained model and implement that model on realtime camera and further image processing stuff...Any reference will be appreciated 👏👏
check out the readme for my "Object Detection API" video or my "How to Build an Object Detection Classifier" video, both will do the trick
Awesome video sir! I am new to your channel and I am already impressed by the videos you make. Thanks for sharing everything!!! 💛💛💛
I always wanted to do deep learning but the resources on the internet were really difficult to understand. Thank you since you really explained it well!
Appreciate it! Thanks for the support.
when i run this cell
# train your custom detector
!./darknet detector train data/obj.data cfg/yolov3_custom.cfg darknet53.conv.74 -dont_show
i am getting this type of error
/bin/bash: ./darknet: Is a directory
please help
Ravirajsinh Dabhi you need to run “%cd darknet” command before, you must be in the root directory and you need to be in the darknet folder to run the detection.
@@TheAIGuy sorry for to trouble you after
i am getting
Couldn't open file: data/obj.data
@The AI Guy ok it's solved thank you for this lovely content.If i want to contribute something and support this content for you what can i do?where can i do?
Ravirajsinh Dabhi just liking this video and subscribing to the channel means a lot to me. Thanks!
@The AI Guy this is my dataset for custom model,this is whole dataset collected by me and my friend from casting foundry
www.kaggle.com/ravirajsinh45/real-life-industrial-dataset-of-casting-product
This is we doing for identify defective product and ok product!! For reducing tedious process of manual inspection of all product.
We already made binary classifier from this dataset now we working for detection of it.
We already made working model of that with help of arduino and stepper motor which can push defective from inspection line.
when i want to test with a video,
instead of this cord:
!./darknet detector test data/obj.data cfg/yolov3_custom.cfg /mydrive/yolov3/backup/yolov3_custom_last.weights /mydrive/images/safari.jpg -thresh 0.3
which command do i have to use??
and do i need another weight file download??
wondering this too
me too!
Yes, please help.
This was an exact video I was looking for..thank u for making it.. saved my time..
One of the best channel for computer vision beginners!!!!
After running the JavaScript code in console it's showing error...
Akhil G Krishnan whats the error
@@TheAIGuy Uncaught TypeError: Cannot read property 'click' of null
at ClickConnect (:3:55)
Uncaught error: ErrorEvent {isTrusted: true, message: "Script error.", filename: "", lineno: 0, colno: 0, …}
@@TheAIGuy Uncaught TypeError: Cannot read property 'click' of null
at ClickConnect (:3:55)
Uncaught error: ErrorEvent {isTrusted: true, message: "Script error.", filename: "", lineno: 0, colno: 0, …}
@@TheAIGuy Uncaught TypeError: Cannot read property 'click' of null
at ClickConnect (:3:55)
Uncaught error: ErrorEvent {isTrusted: true, message: "Script error.", filename: "", lineno: 0, colno: 0, …}
GK MUSIC INDIA try changing the “colab-toolbar-button#connect” to “paper-icon-button” inside the queryselector
nice video.
please make related projects like counting custom objects, vehicle speed predictions etc. your way of teaching and code explanation is absolutely understandable. that is why i'm asking for. thank you.
You are the legend. I used to hate colab because of disconnection with internet. Thanks for your hack and all other tutorial. Liked and Subscribed !
Thanks so much! Glad you found it useful.
LOL! Love the "OH! WHAT'S THAT? I'D DEFINITELY HIT THAT BUTTON". Great video, subscribed.
Hi, I just discovered your channel and loved the content about computer vision. I'll study your videos and many thanks for sharing.
Harry glad you enjoy! Thanks for the kind words. All the best
Thankyou very much for the wonderful video ! It just helped me to sort out all kinda doubts I had related to this topic.
Great video!! Extremely helpful.
Very Nice Tutorial! Thanks for sharing!
you're a fucking legend. I spent a week compiling darknet on my laptop, built my own labelmaker with python and when i got to train my model on cpu it estimated 1600 hours. nearly had a mental breakdown. my first reflex was to look into AWS EC2. spent a day with that headache. then youtube suggests your video and god man you have saved my life. subbed and forever your follower. thank you a thousand
This kinds of abusive words if you praise someone are not good....praise in good words
man I really appreciate it ,this really make trainnig custom object model simple and it helps me a lot.
Appreciate it! Thanks for the kind words. All the best.
Great stuff. Thank you for your hard work. This is very helpful
my task is to detect a paper being help up. it's a bit tricky because most of the test pictures have a white wall background resembling the paper but the network is doing a pretty good job so far, thank you
Thank you so much. Very easy to understand and follow-up
The AI Guy
Grande bro. Im from Chihuahua, Mexico and this really help me
Thanks a lot for your beautiful content sir that helped me build my own custom yolov3 model 🤟🏻
you are amazing man thanks a lot for the tutorial
Thank you so much. From Vietnam with love :3
you're amazing i successfully made my own custom model thank you! :)
Really great tutorial, thanks a lot for sharing knowledge
I am following your toturials for days now and find them extremely clear, thank you for making rgeat content!
Question:
When running a custom training model for one class on my own laptop, for a batch of 4000 it will take 67 hours. How much time do you reckon it will take on this cloud VM?
Might give it a try this weekend, thanks!
Damn good.Keep making more creative videos like this!
This video content helped me a lot... thanks for sharing your knowledge...
Thank You so much!!! Creating custom detector was challenging but it works! Not perfectly but still ;) (It detects malaria cells)
i loved following along,thanks mate
I am making my custom detector on particular traffic sign classes such as stop sign, right-hand side curve etc. As of now to check that how does it work out for only 4 classes, if it works well i will try to train it for more signs around 40 no of classes😅 I hope that won't be an issue in case of that much classes. This tutorial helped a lot. Thanks.🙌
Did you do that does this work
Great video! Thanks!
Glad you enjoyed!
Hello from Brazil, very great tutorial! Congratulations! I am training on BDD100K dataset, it helped a lot! Thank you.
Good explaining 👍👍👍👍
Glad I was looking for this one
Good explanation...keep rocking
you frcking legend, im using this for my master thesis, this just took me several weeks with software with your skript i was able to do in a couple hrs
Haha glad to help. Thanks!
Sitze auch gerade an der MA, kann ich dich irgendwie kontaktieren? :D
Great videos, thanks for sharing, keep it up
Thank you very much...can you continue making a video about test and validate process image?
Thanks a lot man. you should make more and more videos i am hoping to learn tensorflow from you.
Take our subscriptions bro you deserve it!
This is an absoulutly awesome video!! Do you think it the process has changed that much since 2020?
Thanks for the video. I love it
Hi Al Guy, thank you very much for this amazing video.
I need some clarifications on trained weights.
Aftet the training, weights will be saved into the drive as u explained. By any chance, those are those weights downloadable and able to use inside an android application for objects detection?
thank you so much for this!
Sick, thanks so much!
Really amazing videos
Very helpful !!!
Made my day ❤️🔥
Awesome buddy but just one humble request try to upload regularly 2 times per week or maybe 1
Hi, how do we resume a colab project because after training I closed my tabs and when I opened again my darknet folder was no longer there and I had to start from scratch
Thank you for this great explanation. I am just wondering in Yolo with custom data and train, at the end when we give a test image how I can get box location? same as .txt I provided for training.
Thanks
Thanks a lot for this
Thank you man for great tutorial! I have a question though. I was training model for vehicle detection and classification. Recently I found out that vehicles from particular location are not being detected (cuz of some obstacle). I decided to collect more data from that particular location. Thing is that I have already trained on old dataset and got final weights. Do I have to retrain model on whole dataset again (old dataset + new data) or I can just make dataset on new data and begin training from last weights?
Interesting question. I think that the answer for this is transfer learning. You freeze training weights of the old dataset and you retrain with the newer one to update the model.
@@khawlasghir6586 thank you for reply. Yes I've already came up with solution and did what you've just wrtitten. 😁👌
Awesome video! What command do I write to test a video and see the result using the generated weight?
Firstly, thanks for this tutorial, i have been searching tutorial like this for a big while. It seems that it is gonna detect object in images. However, i am wondering that, will it be working for videos as well? Can we use these weights for object detection on video? Thanks in advance for your answer.
Sincerely,
hahhah man. you are so nice... Good one that way to subscribe you, definitely did it!!!
Great video, truly a piece of art, just out of curiosity for how many iterations usually does the training goes?
the standard for high accuracy is usually 2000*(# of classes), so if you are creating a detector for 3 classes you would train for 6000 iterations
It's giving me an error while training, can't open label file(this can be Normal if your MSCOCO)
Is this error normal or should I do some changes for it
@@abuzarshaikh9217 this probably means your obj.names isnt in the right folder or isnt being reference properly in the obj.data folder. I would double check both
Thanks man !
Thank you Brother
Hi. thank you for amazing video. I have one question: How could I omit bounding box labels in test images?? I want to see just bounding box without labels in test images.
Very useful
Hi, thanks for the video! Is there any chance you will do a similar video for yolov3 tiny? Are there major differences or is it about the same thing?
Hi, thank you for this toturial. I have a question, I want to reduce specifics class in yolov3 dataset(coco dataset). In other word, I don't want to detect some classes. How can I do this?
Thank you very much for the video! Is there a way to change the color of the bounding boxes? It would be really important for me to detect each and every object with the color that is defined for them in my school task.
Hi, I would like to test the model with my 20% test data after training. I noticed in the video that when creating the obj.data file you mentioned that we wouldn't be using the test.txt. If I would like to test the model I assume I need to generate the test.txt file. Is there anything else I need to change or do? Will it then provide me with the accuracy of the model after training?
Thank you very much.
Just one thing, my console says- you are connected to a GPU runtime, but not utilizing the GPU.
super thambi
A great tutorial. This really helped me to clear all the doubts regarding YOLOV3 implementation. I have another question.Do you know how to get the coordinates of the bounding box?
ten sor to get bounding boxes just add the flag -ext_output to your detection command
hey your tutorial is super good, I really learned a lot. Can you please tell me how I can store output of each detection in a file along with the coordinates. Thanks
@@hibazafar1815 Hi there and thanks for the kind words! Refer to the official readme of alexeyab/darknet on github as he shows how to do it with detail. A way that works for me is adding this to the end of your darknet detection command '-ext_output >> results.txt'. This will save the detections in a text file. Make sure to then save the results file to your machine or google drive.
Thank you soooooooooooooo much
it is good works and plese share share rhe dolabs all works that you
have done
@TheAIGuy...brilliant tutorial....is there any possible way to save the object detection result into a .txt file on ur local machine (e.g if the predicted class is "dog" then save dog inside a .txt file) ...Thanks
I have one question here. How can we use preprinted model and add our newly trained model with that. For example i have 80 classes in coco names but now i trained model with 2 new classes and i want that 80 and also 2 new classes. So in result my model is able to detect total 82 object. Please Help me for this...
i am getting this error...what should i do??
File "generate_train.py", line 4, in
os.chdir(os.path.join("data", "obj"))
FileNotFoundError: [Errno 2] No such file or directory: 'data/obj'